Professional Documents
Culture Documents
INTRODUCTION
The utilization of the global positioning system (GPS) in photogrammetric mapping began almost from the inception of this technology. Initially, GPS offered a major improvement in the control needed for mapping. It provided coordinate values that were of higher quality and more reliable than those using conventional field surveying techniques. At the same time the cost and labor required for that control were lower than conventional surveying. Experiences from using GPS-control showed several improvements [Salsig and Grissim, 1995]: a)There was a better fit between the control and the aerotriangulation results, particularly for large-area projects. b)Surveyors were not concerned with issues like intervisibility between control points, therefore the photogrammetrist often received the control points in locations advantageous to them instead of the location determined from the execution of a conventional field survey. c)Visibility of the ground control point to the aerial camera is always important. Fortunately, those points that are visible using the GPS receivers are also free of major obstructions that would prevent the image from appearing in the photography. This led to a better recovery rate for the control. Unfortunately, the window from which GPS observations could be made was not always at the most desirable time of day. This changed as the satellite constellation began to reach its current operational status. Also, with these increasing windows came the idea of placing a GPS receiver within the mapping aircraft. Airborne-GPS is now a practical and operational technology that can be used to enhance the efficiency of photogrammetry, although Abdullah et al [2000] reports that only about 30% of the photogrammetry companies are using this technology at this time. But, this does account for about 40% of the projects undertaken by photogrammetric firms. Airborne GPS can be used for: precise navigation during the photo flight, centered or pin-point photography, determination of the coordinates of the nodal point for aerial triangulation To achieve the first two applications the user requires real- time differential GPS positioning [Habib and Novak, 1994]. Because the accuracy of position for navigation and centered photography ranges from one to five meters, C/A-code 1 or P-code2 pseudorange is all that is required. The important capability is the real-time processing. For aerotriangulation, a higher
1
C/A Code: The standard (Clear/Acquisition) GPS PRN code, also known as the Civilian Code or S-Code. Only modulated on the L1 carrier. Used by the GPS receiver to acquire and decode the L1 satellite signal, and from which the L1 pseudo-range measurement is made. 2 P-Code: The Precise or Protected code. A very long sequence of PRN binary biphase modulations on the GPS L1 and L2 carrier at a chip rate of 10.23MHz, which repeats about every 267 days. Each one week segment of this code is unique to a GPS satellite and is reset each week. Only US military and other authorised users are able to overcome AS using special receivers.
accuracy is needed which means observing pseudorange and phase. Here, real-time processing is not as important in terms of functionality. Airborne GPS is used to measure the location of the camera at the instant of exposure. This gives the photogrammetrist XL, YL, and ZL. GPS can also be used to derive the orientation angles by using multiple antennas. Unfortunately, the derived angular relationships only have a precision of about 1 of arc while photogrammetrists need to obtain these values to better than 10 of arc. To compute the position of the camera during the project, two dual frequency geodetic GPS receivers are commonly employed. One is placed over a point whose location is known and the other is mounted on the aircraft. Carrier phase data are collected by both receivers during the flight with sampling rates generally at either 0.5 or 1 second. The integer ambiguity must be taken into account and this will be discussed later. Generally, on-the- fly integer ambiguity resolution techniques are employed.
b)
c)
d)
While airborne GPS can be used to circumvent the necessity of ground control, it offers the
photogrammetrist additional advantages. These include [Abdullah et al, 2000; Lucas, 1994]: It has a stabilizing effect on the geometry. The attainable accuracy meets most mapping standards. Substantial cost reduction for medium and large-scale projects are possible. There is an increase in productivity by decreasing the amount of ground control necessary for a project. It reduces the hazards due to traffic, particularly for highway corridor mapping. Precise flight navigation and pin-point photography are possible with this technology. It is now possible, at least theoretically, to use GPS aerotriangulation without any ground control. This requires [Lucas, 1996] a near perfect system, an unlikely scenario. Moreover, it would be extremely prudent to have control, if for no other reason than to check the results. While airborne GPS is operational, there are special considerations that must be accounted for to ensure success for a project. Airborne GPS is operational and being used for more mapping projects. There are some concerns that need to be addressed for a successful project. These include [Abdullah et al, 2000]: Risk is greater if the project is not properly planned and executed. There is less ground control. As ground control gets smaller, datum transformation problems become more important. There is some initial financial investment by the mapping organization. Requires non-traditional technical support.
ERROR SOURCES
The use of GPS in photogrammetry contains two sets of error sources and the introduction of additional errors inherent in the integration of these two technologies. For precise work, these errors need to be accounted for. Photogrammetric errors include the following: a) Errors associated with the placement of targets. The Texas Department of Transportation has determined that an error of 1 cm can be expected in centering the target over the point [Bains, 1995]. This is based on a 10 cm wide cross target. The main problem is that the center of the target is not precisely defined. Errors inherent in the pug device used to mark control on the diapositives. If the pug is not properly adjusted then the point transfer may locate pass- and tie-points erroneously. Regardless, the process of marking control introduces another source of error into the photogrammetric process. Camera calibration is crucial in determining the distortion parameters of the aerial
b)
c)
camera used in photogrammetry. Bains [1995] has found that the current USGS calibration certificate does not provide the information needed for GPS assisted photogrammetry. Merchant [1992] states that a system calibration is more important with airborne GPS. d) The camera shutter can contain large random variability as to the time the shutter is open. Most of the time, this error source is not that important but if this irregularity is too great, contrast within the image could be lost. The major problem with this non-uniformity is when trying to synchronize the time of exposure to the epoch in which the GPS signal collecting data.
Error sources for GPS are well identified. A loss or disruption of the GPS signal could cause problems in resolving the integer ambiguities and could result in erroneous positioning of the camera location thereby invalidating the project. The GPS error sources include: a) Software problems can cause problems with a GPS mission, particularly in the kinematic mode. Some software cannot resolve cycle slips in a robust fashion, although newer on-the- fly ambiguity resolution software will help. There is also a limitation on the accuracy of different receivers used in the kinematic surveys. Geodetic quality receivers, with 1-2 cm relative accuracy, should be employed for projects where high precision is required. Datum problems. The GPS position is determined in the WGS 84 system whereas the survey coordinates are in some local coordinate system or in NAD 27 coordinates where there is no exact mathematical relationship between systems. Signal interruption. This is critical if continuous tracking is necessary in order to process the GPS signal. Interruption may occur during sharp banking turns through the flight. Geometry of the satellite constellation. Receiver clock drift. Although this error is relatively small, this drift should be accounted for in the processing of GPS observations. Multipath. This is particularly problemsome on surfaces such as the fuselage or on the wings. This error is due to reception of a reflected signal, which represents a delay in the reception time.
b)
c)
d) e)
f)
Errors that can be found in the integration of GPS with the aerial camera and photogrammetry are [Bains, 1995; Merchant, 1992; Lapine, nd]: a) The configuration of airborne GPS implies that the two data collectors are not physically in the same location. The GPS antenna must be located outside and on top of the aircraft to receive the satellite signals. The aerial camera is located within the aircraft and is situated on the bottom of the craft. The separation distance between the antenna and camera (the nodal point) needs to be accurately determined. This distance is found through a calibration process prior to the flight. This value can also
be introduced in the adjustment by constraining the solution or by treating it in the stochastic process. b) Prior to beginning a GPS photogrammetric mission, the height between the ground control point and the antenna needs to be measured. Experience has found that there can be variability in this height based on the quantity of fuel in the aircraft. This problem occurs only when the airborne-GPS system is based on an initialization process when solving for the integer ambiguities. The camera shutter can cause problems as was identified above. The effect of this error creates a time bias. Of concern is the ability to trip the shutter on demand. In the worst case, Merchant [1992] points out that the delay from making the demand for an exposure to the midpoint of the actual exposure could be several seconds. For largescale photography this could cause serious problems because of the turbulent air in the lower atmosphere and the interpretation from the GPS signal to the effective exposure time. Early experiments with the Wild RC10 with an external pulse generator showed wide variability in time between maximum aperture and shutter release [van der Vegt, 1989]. The values ranged from 10-100 msec. Traveling at 100 m/sec, positional errors from 1-10 m could be expected. Interpolation algorithm used to compute the position of the phase center of the antenna. Since the instant of exposure does not coincide with the sampling time in the GPS receiver, an interpolation of the position of the antenna at the instant of exposure must be computed. Different algorithms have varying characteristics, which could introduce error in the position. Related to this uncertainty is the sampling rate used to capture the GPS signal. Too low of a rate will increase the processing whereas too high of a rate will degrade the accuracy of the interpolation model. Radio frequency interference can cause problems, particularly onboard the airplane. A receiver that can filter out this noise should be used. One example receiver is the Trimble 4000 SSI with Super-Trak signal processing which has been used successfully in airborne-GPS [Salsig and Grissim, 1995].
c)
d)
e)
Camera Calibration
One of the weak links in airborne GPS involves the camera calibration. As was pointed out earlier, the traditional camera calibration may not provide the information needed when GPS is used to locate the exposure station. What should be considered is a system calibration whereby the whole process is calibrated and exercised under normal operating conditions [Lapine, 1991; Merchant, 1992]. Because of the complex nature of combining different measurement systems within airborne GPS, two important drawbacks are identified with the traditional component approach to camera calibration [Lapine, 1991]: 1. The environment is different. In the laboratory, calibration can be performed under ideal and controlled conditions, situations that are not possible in practice. This leads to different atmospheric conditions and variations in the noise found in photo measurements.
2.
The effect of correlation between the different components of the total system are not considered.
Traditionally, survey control on the ground had the effect of compensating for residual systematic errors in the photogrammetric process [Lapine, 1991; Merchant, 1992]. This is due to the projective transformation where ground control is transformed into the photo coordinate system. The exposure station coordinates are free parameters that are allowed to float during the adjustment thereby enforcing the collinearity condition. With GPS-observed exposure coordinates, the space position of the nodal point of the camera are fixed and ground coordinates become extrapolated variables. Because of this, calibration of the photogrammetric system under operating conditions becomes critical if high- level accuracy is to be maintained.
When planning, try to find those times when the satellite coverage consists of 6 or more satellites with minimum change in coverage [Abdullah et al, 2000]. Also plan for a PDOP that is less than 3 to ensure optimal geometry. Additionally, one might have to arrive at a compromise between favorable sun angle and favorable satellite availability. Make sure that the GPS receiver has enough memory to store the satellite data. This is particularly true when a static initialization is performed and satellite data is collected from the airport. There may also be some consideration on the amount of sidelap and overlap when the camera is locked down during the flight. This will be important when a combined GPSINS system is used. Finally, a flight management system should be used to precalculate the exposure station locations during the flight The limitations attributed to the loss of lock on the satellite places additional demands on proper planning. These problems can be alleviated to some degree if additional drift parameters are used in the photogrammetric block adjustment.
Antenna Placement
To achieve acceptable results using airborne GPS, it is essential that the offset between the GPS antenna and the perspective center of the camera be accurately known in the image coordinate system (figure 1). The measurement of this offset distance is performed by leveling the aircraft using jack above the wheels. Then, either conventional surveying or close range photogrammetry can be used to determine the actual offset.
Figure 1. GPS Offset For simplicity, the camera can be locked in place during the flight. This helps maintain the geometric relationship of the offset vector. But, the effect is that tilt and crab in the aircraft could result in a loss of coverage on the ground unless more sidelap were accounted for in the planning. If the camera is to be leveled during the flight then the amount of movement should be measured in order to achieve higher accuracy.
The location of the antenna on the aircraft should be carefully considered. Although any point on the top side of the plane could be thought of as a candidate site, two locations can be studied further because of their advantages over other sites. These are on the fuselage directly above the camera and the tip of the vertical stabilizer. The location on the fuselage over the camera has the advantage of aligning the phase center along the optical axis of the camera thereby making the measurement of the offset as well as the mathematical modeling easier [Curry and Schuckman, 1993]. Moreover, the crab angle is hardly affected and the tilt corrections are negligible for large image scale [Abdullah et al, 2000]. The disadvantages are as follows. First, the fuselage location increases the probability of multipath. Second, this location, coupled with the wing placement, may lead to a loss of signal because of shadowing. Antenna shadowing is the blockage of the GPS signal, which could occur during sharp banking turns. Finally, mounting on the fuselage may require special modification of the aircraft by certified airplane mechanics. Placing the antenna on the vertical stabilizer will require more work in determining the offset vector between the antenna and the camera [Curry and Schuckman, 1993]. But once determined, it should not have to be remeasured unless some changes would suggest a remeasurement be undertaken. The advantages are that both multipath and shadowing are less likely to occur. Moreover, the actual installation might be far simpler since many aircraft already have a strobe light on the stabilizer, which could easily be adapted to accommodate an antenna.
Vegt, 1989]. This device will create an electrical pulse when the shutter is at the maximum aperture. Prior to determining the exposure station coordinates, the location of the phase center of the antenna must be interpolated. Since the receiver clock contains a small drift of about 1 s/sec., Lapine [nd] suggests that the position of the antenna be time shifted so that the positions are equally spaced. Several different interpolation models can be employed to determine the trajectory of the aircraft. Some of them include the linear model, polynomial approach, spline function, and quadratic time-dependent polynomial. Some field results found very little difference between these methods [Forlani and Pinto, 1994]. This may have been because they used the GPS receiver PPS (pulse per second) signal to trip the shutter on the aerial camera. This meant that the effective instant of exposure was very close to the GPS time signal.
Figure 2. Shutter release diagram for rotary shutters [from Jacobsen, 1991].
One of the most simplest interpolation models is the linear approach. The assumption is made that the change in trajectory from one epoch to another is linear. Thus, one can write a simple ratio as: i di = (X , Y , Z ) d(X , Y , Z) where: i ?(X,Y,Z) di d(X,Y,Z) = = = = time interval between GPS epochs changes in GPS coordinated between two epochs time difference when the exposure was made within an epoch, and changes in GPS coordinates to the exposure time.
The advantage of this model is its simplicity. On the other hand, it assumes that the change in position is linear which may not be true. Sudden changes in direction are very common at lower altitudes where large scale mapping missions are flown. For example, figure 3 shows a sudden change in the Z-direction during the flight. Assuming a linear change, the location of
the receiver could be considerably different than the actual location during exposure. One alternative would be to decrease the sample interval to, say, 0.5 seconds. This would reduce the effect of the error but increase the number of observations taken and the time to process those data.
Figure 3. Effects of linear interpolation model when the aircraft experiences sudden changes in its trajectory between PG epochs Because of the non- linear nature of the aircraft motion, Jacobsen [1993] suggests that a leastsquares polynomial fitting algorithm be used to determine the space position of the perspective center. By varying the degree of the polynomial and the number of neighbors to be included in the interpolation process, a more realistic trajectory should be obtained. The degree and number of points will depend on the time interval between GPS epochs. The added advantage of this method is that if a cycle slip is experienced, it can be used to estimate better the exposure station coordinates than a linear model. A second order polynomial is used by Lapine to determine the position offset, velocity and acceleration of the aircraft in all three axes. This is done by fitting a curve to a five epoch period around the exposure time. The effect of this polynomial is to smooth the trajectory of the aircraft over the five epochs. The following model is used: X1 = a X + b X (t 1 t 3 ) + c X (t 1 t 3 )2 X 2 = a X + b X (t 2 t 3 ) + c X ( t 2 t 3 )
2
X 3 = a X + b X (t 3 t 3 ) + c X (t 3 t 3 )2 X 4 = a X + b X (t 4 t 3 ) + c X ( t 4 t 3 )2 X 5 = a X + bX (t 5 t 3 ) + c X ( t 5 t 3 )
2
Similar equation can be generated for Y and Z. Thus the three models look, in a general form, like:
X = aX + bX t + cX t 2 Y = aY + bY t + cY t 2 Z = a Z + b Z t + c Zt 2
where:
t a b c
= = = =
ti - t3 and i = 1, 2, ..., 5 distance from the origin velocity, and twice the acceleration
The design or coefficient matrix is found by differentiating the model with respect to the unknown parameters. All three models have the same coefficient matrix:
1 1 B = 1 1 1
2 ( t1 t 3 ) (t 1 t 3 ) 1 2 (t 2 t 3 ) (t 2 t 3 ) 1 2 (t 3 t 3 ) (t 3 t 3 ) = 1 2 (t 4 t 3 ) (t 4 t 3 ) 1 2 (t 5 t 3 ) ( t 5 t 3 ) 1
t t2 t t2 t t2 t t2 t t2
X1 X 2 fX = X 3 X 4 X 5
Y1 Y 2 f Y = Y3 Y4 Y5
Z1 Z 2 f Z = Z3 Z 4 Z5
The normal equations can then be expressed as v X = B X + f X v Y = B Y + f Y v Z = B Z + f Z where ? represent the parameters ( ? = [a b c]T ). The solution becomes
x = BT WB Y Z
T T
( ) = (B WB) = (B WB)
1 1
BT Wf X BT Wf Y BT Wf Z
where W is the weight matrix. Assuming a weight of 1, the weight matrix then becomes the identity matrix and
5 T T B WB = B IB = 5t 5 t 2 5t 5t 5t 3
2 2 5t 3 5t 5 t4
X1 + X2 + X3 + X 4 + X5 X t +X t+ X t + X t +X t B Wf X = B If X = 1 2 3 5 2 2 2 2 X1t + X2 t + X3 t + X4 t + X 5t 2
T T
The weighting scheme is important in the adjustment because an inappropriate choice of weights may biased or unduly influence the results. Lapine looked at assigning equal weights but this choice was rejected because the trajectory of the aircraft may be non-uniform. The final weighting scheme us ed a binomial expansion technique whereby times further from the central time epoch (t3 ) were weighted less than those closest to the middle. Using a variance of 1.0 cm2 for the central time epoch, the variance scheme looks like 2 2 0.01 m 2 2 2 2 0.01 m 2 4 cm
2 0.01 m
2
22 0.01 m 2 22 0.01 m 2
4 cm 2 =
4 cm
4 cm 2 4 cm 2
where the off-diagonal values are all zero (0). A basic assumption made in Lapine's study was that the observations are independent therefore there is no covariance. Once the
coefficients are solved for, the position of the antenna phase center can be computed using the following expressions
The use of on-the-fly (OTF) ambiguity integer resolution makes the process much easier. The new GPS receiver and post-processing software are much more robust and easy to use while the receiver is in flight. OTF requires P-code receivers where carrier phase data are collected using both the L1 and L2 frequencies. The solution requires about 10-15 minutes of
1
The unknown number of whole wavelengths of the carrier signal contained in an unbroken set of measurements from a single satellite at a single receiver.
measurements before entering the project area. Component integration can also create problems. For example, a test conducted by the National Land Survey, Sweden, experienced cycle slips when using the aircraft communication transmitter [Jonsson and Jivall, 1990]. Receiving information was not a problem, just transmissions. This test involved pre- flight initialization with the goal of re-observation over the reference station at the end of the mission. This was not possible.
GPS-Aided Navigation
One of the exciting applications of airborne-GPS is its utilization of in flight navigation. The ability to precisely locate the exposure station and activate the shutter at a predetermined interval along the flight line is beneficial for centering the photography over a geographic region, such as in quad-centered photography for orthophoto production. An early test by the Swedish Nationa l Land Survey [Jonsson and Jivall, 1990] showed early progress in this endeavor. The system configuration is shown in figure 4. Two personal computers (PCs) where used in the early test - one for navigation and the other for determination of the exposure time.
Figure 4. Configuration of navigation-mode GPS equipment [from Jonsson and Jivall, 1990]. The test consisted of orientation of the receiver on the plane prior to the mission over a ground reference mark. This initialization is performed to solve for the integer ambiguity. This method of fixing the ambiguity requires no loss of lock during the flight thus necessitating long banking turns, which adds to the amount of data collected. A flight plan was computed with the location of each exposure station identified. The PC used for the navigation activated a pulse that was sent to the aerial camera to trip the shutter. The test showed that this approach yielded about a 0.5 second delay. Thus, the exposure station
locations were 20-40 meters too late. An accuracy of about 6 meters was found at the preselected position along the strip. When compared to the photogrammetrically derived exposure station coordinates, the relative carrier phase measurements were within about 0.15 meters in agreement. The Texas Department of Transportation (TDOT) had a different problem [Bains, 1992]. Using airborne-GPS gave TDOT the ability to reduce the amount of ground control for their design mapping. With GPS one paneled control point was placed at the beginning of the project and a second at the end. If the site was greater than 10 km in length then a third paneled control point was placed near the center. For their low altitude flights (photo scale of 1 cm = 30 m), the desire was to control the side- lap to 50 m. Using real-time differential GPS, accuracies of better than 10 m, at that time, were realistic. Using this 10 m error value, this amount of error would only cause a variation in side- lap of 7%. TDOT uses 60% side- lap for their large scale mapping. For the high altitude mapping (photo scale of 1 cm = 300 m) and 30% side- lap, it was determined that the "50 m was not really necessary. This 50 m value would cause a variation of only about 2%.
where:
xij, yij are the observed photo coordinates, i, for photo j xo , yo are the coordinates of the principal point c is the camera constant ? Xi, ? Yi, ? Zi are the transformed ground coordinates This mathematical model is often presented in the fo llowing form:
x ij + v x ij = x o c m 11 (X i X L ) + m 12 (Yi YL ) + m 13 (Z i Z L ) m 31 (X i X L ) + m 32 (Yi YL ) + m 33 (Z i Z L ) m 21 (X i X L ) + m 22 (Yi YL ) + m 23 (Z i Z L ) m 31(X i X L ) + m 32 (Yi YL ) + m 33 (Z i Z L )
y ij + v yij = y o c
where:
are the residuals in x and y respectively for point i on photo j are the ground coordinates of point i are the space rectangular coordinates of the exposure station for photo j is the 3x3 rotation matrix that transforms the ground coordinates to a photo parallel system.
The model implies that the difference between the observed photo coordinates, corrected for the location of the principal point, should equal the predicted values of the photo coordinates based upon the current estimates of the parameters. These parameters include the location of the exposure station and the orientation of the photo at the instant of exposure. The former values could be observed quantities from onboard GPS. These central projective equations form the basis for the aerotriangulation. It is common to treat observations as stochastic variables. This is done by expanding the mathematical model. For example, Merchant [1973] gives the additional mathematical model when observations are made on the exterior orientation elements as: F( ) = = 0 F() = = 0 F() = = 0 F(X L ) = XL XL = 0 F(YL ) = YL YL = 0 F(Z L ) = ZL ZL = 0
o a o a o a o a o a o a
Figure 5. Position ambiguity for a single photo resection [from Lucas, 1996, p.125]. Using GPS to determine the exposure station coordinates without ground control is not applicable to all photogrammetric problems. Ground control is needed for a single photo resection and orientation [Lucas, 1996]. If the exposure station coordinates are precisely known then the only thing known is that the camera lies in some sphere with a radius equal to the offset distance from the GPS antenna to the cameras nodal point, figure 5. The antenna is located at the center of the circle. All positions on the sphere are theoretically possible but from a practical viewpoint, one knows that the camera, being located below the aircraft and pointing to the ground, is below the antenna. The antenna, naturally, is located on top of the aircraft to receive the satellite signals. Adding a second photo reduces some of the uncertainty. This is due to the additional constraint of the collinearity condition that is placed on the rays from the control to the image position. The collinearity theory will provide the relative orientation between the two photos [Lucas, 1996]. Without ground control, the camera is then free to rotate about a line that passes through the two antenna locations (see figure 6). Without ground control, or some other mechanism to constrain the roll angle along the single strip, this situation could be found throughout a single strip of photography.
Figure 6. Ambiguity of the camera position for a pair of aerial photos [from Lucas, 1996, p.125].
While independent model triangulation continues to be employed in practice, the usual iterative adjustment cannot be used with the recommended 4 corner control points [Jacobsen, 1993]. Moreover, the 7-parameter solution to independent model triangulation results in a loss of accuracy in the solution. Determining the coordinates of the exposure stations can be easily visualized in the following model [Merchant, 1992]. Assume that the photo coordinate system (x,y,z) are aligned with the coordinate system (U, V, W). Further, assume that the survey control (X, Y, Z) is reported in the WGS 84 system. Then, it remains to transform the offset between the receivers phase center and the nodal point of the aerial camera (DU, DV, DW) into the corresponding survey coordinate system. This is shown as
X L Xa Y = Y + M M E M L a Z L ZA
DU DV DW
where:
DU, DV, DW are the offset distances MM is the camera mount orientation ME is the exterior orientation elements of the camera
The camera mount orientation is necessary to ensure that the camera is heading correctly
down the flight path. In the normal acquisition of aerial photos, the camera is leveled prior to each exposure. This is done so that the photography can be nearly vertical at the instant of exposure even though the aircraft is experiencing pitch, roll and swing (crab or drift). When the coordinate offsets between the antenna and camera were surveyed, the orientation angles on the mount are leveled. A problem occurs if there is an offset between the location of the nodal point and the gimbals rotational center on the mount. When the camera is rotated, the relationship between the two points should be considered. The simplest way to ensure that the relationship between the receiver and the camera are consistent would be to forgo any rotation of the camera during the flight. With this rigid relationship fixed, the antenna coordinates can be rotated into a parallel system with respect to the ground by using the tilts experienced during the flight. Alternatively, Lapine [nd] points out that the transformation of the offsets to the local coordinate system can easily be performed using the standard gimbal form. In this situation, pitch and swing angles between the aircraft and the camera are measured. Then, one can simply algebraically sum the camera mount angles with the appropriate measured pitch and swing angles. Here, ? and swing are added to form one rotational element and f and pitch are similarly combined. Since roll was not measured during the test, ? is treated independently. Using the Wild RC-10 camera mount, Lapine found that the optical axis of the camera coincided with the vertical axis of the mount. That meant that the combination of ? and swing would not produce any eccentricity. Testing revealed that the gimbal center was located approximately 27 mm from the nodal points. Thus, an eccentricity error could be introduced. During the flight, a 1.5o maximum pitch angle between the aircraft and the camera mount was found. Thus the error in neglecting this effect in the flight direction would be maximum pitch error = 0.027m * sin 1.5o = 0.0007 meters Experiences from tests in Europe [Jacobsen, 1991] indicate that the GPS positions of the projection centers differ from the coordinates obtained from a bundle adjustment. Moreover, many of the data sets have shown a time dependent drift pattern in the GPS values. When this systematic tendency is accounted for in the adjustment, excellent results are possible. For relative positioning, 4 cm can be reached whereas 60 cm are possible using absolute positioning. A second approach to perform airborne GPS aerial triangulation is sometimes referred to as the Stuttgart method. In this technique, certain physical conditions are assumed or accepted [Ackermann, 1993]. First, it is accepted that loss of lock will occur. This means that low banking angles onboard the aircraft will not be used as in those methods where a loss of lock means a thwarted mission. Because loss of lock, it is also unnecessary to perform a stationary observation prior to take-off to resolve the integer ambiguities. These ambiguities are solved on-the- fly and can be determined for each strip if loss of lock occurs during the banking (or at other times during the photo mission). Seldom will loss of lock happen along the strip though. Second, it will be assumed that single frequency receivers will be used on the aircraft. Finally, the ground or base receiver will probably be located at a great distance from the photo
mission. The solution of the integer ambiguities is performed using C/A-code pseudorange positioning. These positions can be affected by selective availability (SA). Because of this, there will be bias in the solution. These drift errors, which can include other effects such as datum effects, are systematic in nature and consist of a linear and time dependent component. The block adjustment is used to solve for these biases. Early test results added confusion to the drift error biases. In a test by the Survey Department of Rijkswaterstaat, Netherlands, a systematic effect was not noticeable on all photo strips [van der Vegt, 1989]. Evaluation of the results indicated that this was probably due to the GPS processing of the cycle slips. The accuracy of the position in the differential mode is predicated on the accuracy of solving these integer ambiguities at both the base receiver and the rover. This test used a technique where the differences between the observed pseudoranges and the phase measurements were averaged. The accuracy of this approach will be dependent upon the accuracy of the measurements, the satellite geometry and how many uncorrelated observations are used in the averaging approach. If no loss of lock occurs during the photo mission, the aircraft trajectories will be continuous and, therefore, only one set of drift parame ters need to be carried in the bundle adjustment. Unfortunately, banking turns could have an adverse effect by blocking the signal to some of the satellites causing cycle slips. Hghlen [1993] states that an alternative to the strip-wise application of the biases, the block may be able to be split into parts where the aircraft trajectories are continuous thereby decreasing the number of unknown parameters within the adjustment. The advantage of modeling these drift parameters is that the ground receiver does not have to be situated near the site. It could be 500 km or farther away [Ackermann, 1993]. This is important because it can decrease the costs associated with photo missions. Logistical concerns include not only the deployment of the aircraft but also the ground personnel on the site to operate the base. When projects are located at great distances from the airplanes home base, uncertainty in weather could mean field crews already on the site but the photo mission canceled. It also is an asset to flight planning in that on-site GPS ground receivers will require fixing the flight lines at least one day before the mission. During the flying season this could be a problem [Jacobsen, 1994]. In Germany, the problem is solved because of the existence of permanent reference stations throughout the country that could be occupied by the ground receiver. Using the mathematical model for additional stochastic observations within the adjustment as outlined earlier [Merchant, 1973], a new set of observations can be written for the perspective center coordinates as [Blankenberg, 1992].
= the perspective center coordinates observed with GPS = residuals on the observed principal center coordinates = the adjusted perspective center coordinates used within the bundle adjustment
Figure 7. Geometry of the GPS antenna with respect to the aerial camera (zerox copy, source unknown) As it was discussed earlier, the antenna does not occupy the same location as the camera nodal point. The geometry is shown in figure 7. Relating the antenna offset to the ground is dependent upon the rotation of the camera with respect to the aircraft and the orientation of the aircraft to the ground. The bundle adjustment can be used to correct the camera offset if the camera remains fixed to the aircraft during the photo missio n. If this condition is met then the orientation of the camera offset will only be dependent upon the orientation elements (, , ). The new additional observation equations to the collinear model are given as [Ackermann, 1993; Hghlen, 1993]:
x A a X bX PC A y PC + a Y + dt b Y z A a Z b Z j PC j
(XA, YA, ZA)GPS = ground coordinates of the GPS antenna for photo i vX, vY, vZ = residuals for the GPS antenna coordinates (XA, YA, ZA)GPS for photo i XL, YL, ZL = exposure station coordinates of photo i xAPC, yAPC, zAPC = eccentricity components to the GPS antenna aX, aY, aZ = GPS drift parameters for strip j representing the constant term dt = difference between the exposure time for photo I and the time at the start of strip j
= GPS drift parameters for strip j representing the lnear timei dependent terms = orthogonal rotation matrix.
It is recognized in analytical photogrammetry that adding parameters to the adjustment weakens the solution. To strengthen the problem, one can introduce more ground control, but this defeats one of the advantages of airborne GPS. Introducing the stepwise drift parameters and using four ground control points located at the corners of the project, there are three approaches to reducing the instability of the block [Ackermann, 1993]. These are shown in figure 8 and are: i) ii) iii) using both 60% end- and 60% side- lap using 60% end- lap and 20% side- lap and adding an additional vertical control point at both ends of each strip, and using the conventional amount of overlap as indicated in (ii) and flying at least two cross-strips of photography.
The block schemes shown in figure 8 are idealized depictions. The figure 8(i) scheme can be
Figure 8. Idealized block schemes. used for airborne GPS when no drift parameters are employed in the block adjustment. It is important that the receiver maintains lock during the flight which necessitates flat turns between flights. Maintaining lock ensures that the phase history is recorded from take-off to landing. Abdullah et al [2000] points out that this is the most accurate type of configuration in a production environment. The same control scheme can also be used when block drift parameters are used in the bundle adjustment. If strip drift parameters are used then a control configuration as shown in figure 8(ii) should be used. Here, drift parameters are developed for each flight line strip which requires additional height control at the ends of each strip. The control configuration in figure 8(iii) incorporates two cross strips of photography. This model increases the geometry and provides a check against any gross errors in the ground control. But it does add to the cost of
the project because more photography is required to be taken and measured. For that reason, it is not frequently utilized in a production environment. More often the area is not rectangular but rather irregular. In this situation it is advisable to add additional cross-strips or provide more ground control. Figure 9 is an example.
Figure 9. GPS block control configuration. Theoretically, it is possible to perform the block adjustment without any ground control. This can easily be visualized if one considers supplanting the ground control by control located at the exposure stations. Nonetheless, it is prudent to include control on every job, if nothing more than providing a check to the aerotriangulation. Using the four control point scheme as just presented has the advantage of using the GPS position for interpolation only within the strip. As is known, conventional aerotriangulation requires ground control. As an example, for planimetric mapping, control is required at an interval of approximately every seventh photo on the edge of the block. Topographic mapping requires vertical control within the block at about the same spacing. Using this background and simulated data, Lucas [1996] was able to
develop error ellipses from a bundle adjustment showing the accumulation of error along the edges of the block (figure 10). The is commonly referred to as the edge effect and stems from a weakened geometric configuration that exists because of a loss in redundancy. Under normal circumstances, a point in the middle of a block should be visible on at least nine photos. But on the edge, the photos are taken only from one side of view.
Figure 10. Error ellipses with ground points positioned by conventional aerotriangulation adjustment of a photo block [Lucasm 1994]. Using the same simulated data, Lucas [1996] also showed the error ellipses one would expect to find using 60% end- and side- lap photography along with airborne GPS and no control. The results show that for planimetry, the results are similar. Larger error ellipses were found at the control points but at every other point they were either smaller or nearly equivalent. Elevation errors were much different within the two simulations. Using just aerotriangulation without control, error ellipses grew larger towards the center of the block. Using kinematic GPS, on the other hand, kept the error from getting larger. Compared with the original simulation with vertical control within the block, each point had improvements, except the control points that were fixed in the conventional adjustment. Lucas [1996] states that the reason for the improvement lies in the fact that each exposure station is now a control point and the distance between the control is less than one would find conventionally. It would not be practical to have the same density of control as one would have in the air. These results are based on simulations therefore reflect what is possible and not necessarily what one would find in real data. Accuracy considerations are important in determining the viability of using GPS observations within a combined bundle adjustment. Results of projects conducted with combined GPS bundle adjustment show that this approach is not only feasible but also desirable. In conventional aerotriangulation, ground control points helped suppress the effects of block
deformation. GPS observed perspective center coordinates stabilize the adjustment thus negating the necessity for extensive control. In fact, their main function now becomes one of assisting in the datum transformation problem [Ackermann, 1993]. If the position of exposure station can be ascertained to an accuracy of 10 cm or better, then the accuracy of the adjustment becomes primarily dependent upon the precision of the measurement of the photo coordinates [Ackermann, 1993]. Designating the standard error of the photo observations as s 0 , the projected values expressed in ground units are O . Then as long as GPS O , Ackermann indicates that the following rule could apply. The expected horizontal accuracy (X, Y) will be approximately 1.5 O and the vertical accuracy (Z) would be around 2.0 O . This assumes using the six drift parameters for each strip, four control points and cross-strips.
Figure 11
Figure 12. The middle photo had 30 targeted image points. For this test, only one or two were used while the remaining control values were withheld. The results are shown in the following table. The full field method utilized all of the checkpoints within the photography. The corridor method only used a narrow band of points along the route, which is typical of the area of interest for many transportation departments [Merchant, 1994]. The results are expressed in terms of the root mean square error (rmse) defined as the measure of variability of the observed and "true" (or withheld) values for the checkpoints. The method is shown as:
rmse = ( true observed ) n
2
rmse (meters) X Y Z
Using 2 targeted ground control points Full Field Corridor 28 10 0.079 0.031 0.057 0.026 0.087 0.073
Using 1 targeted ground control point Full Field Corridor 29 11 0.084 0.034 0.050 0.033 0.086 0.082
The results indicate that accuracies in elevation are better than 1:20,000 of the flying height, which are comparable to results found from conventional block adjustments. It should also be noted that pass points were targeted therefore errors that may occur due to the marking of conjugate imagery is not present. Moreover, the adjustment also included calibration of the system. Nonetheless, good results can be expected by using ground control to alleviate the ill conditioning of the normal equations. A minimum of one point is needed with additional points being used as a check. Another approach, other than including additional control, would be to fly a cross strip perpendicular to the strip of photography. This will have the effect of anchoring the strip thereby preventing it from accumulating large amounts of error. If the strip was only a single strip, then it is recommended that a cross strip be obtained at both ends of the strip [Lucas, 1996].
The mission began by measuring the height of the antenna when the aircraft was parked. The ground receiver was turned on and a sample rate of 1 second was used. The rover receiver in the aircraft was then turned on and tracked the satellites for five minutes with the same one-second sampling rate. Then the aircraft took off and flew its mission. The processing steps involved the kinematic solution of the GPS observations. The PNAV software was used for on-the- fly ambiguity resolution. The software vendor recommended that the processing be done both forward and backward for better accuracy but the test indicated that, at least for this project, there was no increase in the accuracy when performing that kind of processing. The photogrammetry was processed using soft-copy photogrammetry. A 15Fm pixel size was used. The aerial triangulation was then performed with the GAPP software using only four ground control stations; two at the start and two at the end. The results were then statistically processed using the SAS (Statistical Analysis System). The results of this study showed that the accuracy achieved fell within specifications. In fact, the GPS results were either equal to or better than the accuracy of conventional positioning systems. The results also indicated that there was a need to have a reference point within the site to aid in the transformation to State Plane Coordinates. As an example, Table 1 shows the comparison between the GPS-derived control and the values from the ground truth. These results show that airborne GPS can meet the accuracy specifications for photogrammetric mapping.
Number of Observations 95 95 95
Variable
Minimum
Maximum
Mean
Table 1. Comparison of airborne GPS assisted triangulation with ground truth on day 279, 1993 over a long strip [from Bains, 1995, p.40].
ECONOMICS OF AIRBORNE-GPS
While no studies have been conducted that describe the economic advantages of airborneGPS, some general findings are available [Ackermann, 1993]. Utilization of airborne-GPS does increase the aerotriangulation costs by about 25% over the conventional approach. This increase includes: flying additional cross-strips film GPS equipment GPS base observations processing the GPS data and computation of aircraft trajectories aerotriangulation point transfer and photo observations, and combined block adjustment
The real savings accrue in the control where the costs are 10% or less than those required using conventional aerotriangulation. The overall net savings will be about 40% when looking at the total project costs. If higher order accuracy is required (Ackermann uses the example of cadastral photogrammetry which needs 1-2 cm accuracy) then the savings will decrease because additional ground control are necessary.
REFERENCES
Abdullah, Q., M. Hussain and R. Munjy, 2000. Airborne GPS-Controlled AerialTriangulation: Theory & Practical Concepts, Workshop notes. Ackermann, F., 1993. "GPS for Photogrammetry", The Photogrammetric Journal of Finland,
13(20):7-15. Bains, H.S., 1992. "Photogrammetric Surveying by GPS Navigation", Proceedings of the 6th International geodetic Symposium on Satellite Positioning, Vol. II, Columbus, OH, March 1720, pp 731-738. Bains, H.S., 1995. "Airborne GPS Performance on a Texas Project", ACSM/ASPRS Annual Convention and Exposition Technical Papers, Vol. 2, February 27 - March 2, pp 31-42. Corbett, S.J. and T.M. Short, 1995. "Development of an Airborne Positioning System", Photogrammetric Record, 15(85):3-15. Curry, S. and K. Schuckman, 1993. Practical Guidelines for the Use of GPS Photogrammetry, ACSM/ASPRS Annual Convention and Exposition Technical Papers, Vol. 3, New Orleans, LA, pp 79-88. Forlani, G. and L. Pinto, 1994. "Experiences of Combined Block Adjustment with GPS Data", International Archives of Photogrammetry and Remote Sensing, Vol. 30, Part 3, Munich, Germany, September 5-9, pp 219-226. Habib, A. and K. Novak, 1994. "GPS Controlled Aerial Triangulation of Single Flight Lines", Proceedings of ASPRS/ACSM Annual Convention and Exposition, Vol 1, Reno, NV, April 25-28, pp 225-235; also, International Archives of Photogrammetry and Remote Sensing, Vol. 30, Part 2, Ottawa, Canada, June 6-10, pp 203-210. Hghlen, A., 1993. "GPS-Supported Aerotriangulation in Finland - The Eura Block", The Photogrammetric Journal of Finland, 13(2):68-77. Jacobsen, K., 1991. "Trends in GPS Photogrammetry", Proceedings of ACSM-ASPRS Annua l Convention, Vol. 5, Baltimore, MD pp 208-217. Jacobsen, K., 1993. Correction of GPS Antenna Position for Combined Block Adjustment, ACSM/ASPRS Annual Convention and Exposition Technical Papes, Vol. 3, New Orleans, LA pp 152-158. Jacobsen, K., 1994. Combined Block Adjustment with Precise Differential GPS-Data, International Archives of Photogrammetry and Remote Sensing, Vol. 30, Part 3, Munich, Germany, September 5-9, pp 422-426. Jonsson, B. and A. Jivall, 1990. Experiences from Kinematic GPS Measurements, Paper presented at the Nordic Geodetic Commission 11th General Meeting, Copenhagen, 12p. Lapine, L.A., 1991. Analytical Calibration of the Airborne Photogrammetric System Using A Priori Knowledge of the Exposure Station Obtained from Kinematic Global Positioning System Techniques, Department of Geodetic Science and Survey Report No. 411, The Ohio State University, Columbus, OH, 188p. Lapine, L.A., nd. "Airborne Kinematic GPS Positioning for Photogrammetry - The
Determination of the Camera Exposure Station", Xerox copy, source unknown. Lucas, J.R., 1996. Covariance Propagation in Kinematic GPS Photogrammetry in Digital Photogrammetry: An Addendum to the Manual of Photogrammetry, ASPRS, pp 124-129. Merchant, D.C., 1992. "GPS-Controlled Aerial Photogrammetry", ASPRS/ACSM/RT92 Technical Papers, Col. 2, Washington, D.D., August 3-8, pp 76-85. Merchant, D.C., 1994. "Airborne GPS-Photogrammetry for Transportation Systems", Proceedings of ASPRS/ACSM Annual Convention and Exposition, Vol. 1, Reno, NV, April 25-28, pp 392-395. Salsig, G. and T. Grissim, 1995. GPS in Aerial Mapping, Proceedings of Trimble Surveying and Mapping Users Conference, Santa Clara, CA, August 9-11, pp 48-53. van der Vegt, 1989. Differential GPS: Efficient Tool in Photogrammetry, Surveying Engineering, 115(3):285-296.