Professional Documents
Culture Documents
Abstract: This work presents basic methods in least squares adjustment computation. These methods are first principles technique,
observation equations and condition equations techniques. A simple numerical example is used to elucidate these basic methods.
Including experimenting other more recent methods of adjustment such as: least squares collocation, Kalman filter and total least
squares.
Keywords: Least squares, least squares collocation, Kalman filter, total least squares, adjustment computation
1. Introduction
La = f ( X a )
Where,
La = Lb + V
X a = X o + X
PV
i =1
is minimum
(1)
Where,
P is weight of observations, V is the residual and
number of observations. V is expressed as,
Vi = yi y i
n is the
(2)
Where,
y represents the original observations
(4)
(5)
Where,
determined
Lb + V = f ( X o + X )
(3)
(6)
1988
Lb + V = f ( X o ) +
Assuming,
A=
and
f ( X o )
. X + ...
X a
(7)
f ( X o )
X a
o
(8)
L = f (X ) L
(9)
rewritten as,
V = AX + L
(10)
= V T PV K T ( AX + L V )
is minimum
(11)
T
Differentiating equation 11 partially with respect to V , K
and X ,
= PV T + K T = 0
(12)
= AX L + V = 0
T
(13)
= K T A = 0
(14)
X = ( A PA) A PL
1
(15)
(16)
(17)
f ( Lb + V ) = 0
(20)
f ( Lb ) +
f ( Lb )
V + ... = 0
La
(21)
Assuming,
W = f ( Lb )
(22)
and
B=
f ( Lb )
La
(23)
= V T PV K T ( BV + W ) is minimum
(25)
V and
= V T P K T B = 0
(26)
= BV W = 0
K T
(27)
La = Lb + V , therefore equation 19
K ,
L = L
Recall equation 4,
becomes,
(19)
X = ( AT PA) 1 AT PLb
f ( La ) = 0
f (X o ) = 0
BV + W = 0
V = P 1 B T ( BP 1 B T ) 1W
(28)
1989
= 2( X 1a 11.152) + 2( X 1a + X 2a 24.684) +
a
X 1
2( X 1a + X 2a + X 3a 36.711) = 0
3. Application
A short-range Electronic Distance Measurement (EDM)
instrument was used to measure the distances shown in
Figure 1 and Table 1 below, along a straight baseline design
[1]. It was assumed that the measurements were equally
weighted; therefore P = I .
(36)
= 2( X 2a 13.499) + 2( X 1a + X 2a 24.684) +
X 2a
2( X 2a + X 3a 25.539) +
2( X 1a + X 2a + X 3a 36.711) = 0
(37)
= 2( X 3a 12.052) + 2( X 2a + X 3a 25.539) +
X 3a
Figure 1: Baseline measurements
2( X 1a + X 2a + X 3a 36.711) = 0
AB
BC
CD
AC
BD
AD
DISTANCES (m)
11.152
13.499
12.052
24.684
25.539
36.711
(38)
6 X 1a + 4 X 2a + 2 X 3a = 145.094
(39)
4 X 1a + 8 X 2a + 4 X 3a = 200.866
(40)
2 X 1a + 4 X 2a + 6 X 3a = 148.604
(41)
X 1a 11.165m
a
X 2 = 13.504m
X 3a 12.043m
(42)
P1V12 = 1( X 1a 11.152) 2
(29)
P2V22 = 1( X 2a 13.449) 2
(30)
P3V32 = 1( X 3a 12.052) 2
(31)
P4V42 = 1( X 1a + X 2a 24.684) 2
(32)
L1a = X 1a
(43)
P5V52 = 1( X 2a + X 3a 25.539) 2
(33)
La2 = X 2a
(44)
P6V62 = 1( X 1a + X 2a + X 3a 36.711) 2
(34)
La3 = X 3a
(45)
La4 = X 1a + X 2a
(46)
La5 = X 2a + X 3a
(47)
La6 = X 1a + X 2a + X 3a
(48)
PV
i
i =1
therefore,
is minimum;
a
2
a
3
( X 1a + X 2a 24.684) 2 + ( X 2a + X 3a 25.539) 2 +
a
1
a
2
a
3
( X + X + X 36.711) is minimum
(35)
Recall that
X = ( AT PA) 1 AT PLb
1990
X 1a
X = X 2a ,
X 3a
Recall
(56)
from
equation
1 T
1 T 1
that V = P B ( BP B ) W . Also
a
b
equation 4 that L = L + V , therefore,
1
`0
0
A=
1
0
0 0
1 0
0 1
,
1 0
1 1
1 1
1
0
0
P=
0
0
0 0 0 0 0
1 0 0 0 0
0 1 0 0 0
and
0 0 1 0 0
0 0 0 1 0
0 0 0 0 1
(50)
(51)
(52)
1
0
Recall that
0
P=
0
0
X 1a 11.165m
a
X 2 = 13.504m
X a 12.043m
(53)
from
0
0
0
0
0
V yielded,
0.013
0.005
0.009
V =
0.015
0.008
0.001
Recalling
therefore,
recall
1 1 0 1 0 0
B = 0 1 1 0 1 0 and
(57)
1 1 1 0 0 1
11.152+13.499 24.684 0.033
W = 13.499+12.052 25.539 = 0.012 (58)
11.152+13.499+12.052 36.711 0.008
11.152
13.499
12
.
052
Lb =
24.684
25.539
36.711
28
(59)
11.165m
13.504m
m
12
.
043
La = Lb + V =
24.670m
25.547m
36.712m
(60)
(54)
Y = AX + Z
(61)
(55)
Where Z = R + S
(62)
1991
R and S are
given as,
1
1
X = ( AT Y A) 1 AT Y Y [5, 10]
(63)
X k = AX k 1 + wk 1
(64)
Z k = HX k + v k
(65)
Where X k is estimated state at time k , A is the state
transition matrix, X k 1 is estimated state for preceding time
k 1 , w is process noise at time k 1 , Z k is the
measurement, H is the measurement design matrix and v k
is the measurement noise [3].
Figure 2: Residuals
The computed residuals for each observation were given in
Figure 2. From Figure 2, similar residual values were yielded
all the four models. When the sum of the absolute values of
the residuals was computed for the four models (Figure 3), it
was found that KF yield the least value, while the remaining
three models yielded relatively the same values.
For TLS, it assumes that all the elements of the data are
erroneous. This situation can be stated mathematically as,
b + b = ( A + A) x , rank ( A) = m < n
(66)
where, b is error vector of observations and A is error
matrix of data matrix A . Both errors are assumed
independently and identically distributed with zero mean and
with same variance [9].
Table 2: Adjusted distances
a
1
X 2a
X
a
3
LS
11.165
2
13.504
2
12.042
8
LSC
11.165
2
13.504
2
12.042
8
KF
11.165
0
13.504
2
12.042
5
TLS
11.165
3
13.504
3
12.042
8
X 1a , X 2a and
4. Conclusion
The solutions from the first principles, observation equations
and condition equations for
1992
References
[1] O. Okwuashi, Adjustment computation and statistical
methods in surveying, A Manual, in the Department of
Geoinformatics & Surveying, Faculty of Environmental
Studies, University of Uyo, Nigeria, 2014.
[2] C. F. Gauss, Theoria combinationis obsevationum
erroribus minimis obnoxiae, Werke, Vol. 4. Gttingen,
Germany, 1823.
[3] J. Bezrucka, The use of a Kalman filter in geodesy and
navigation, Slovak Journal of Civil Engineering, Vol.
2, pp. 8-15, 2011.
[4] R. E. Kalman, A New Approach to Linear Filtering
and Prediction Problems, Transactions of the ASMEJournal of Basic Engineering, (Series D), Vol. 82, pp.
35-45, 1960.
[5] H. Moritz, Least squares collocation, Reviews of
Geophysics and Space Physics, vol. 16, No. 3, pp 421430, 1978.
[6] G. H. Golub, C. F. V. Loan, An analysis of the total
least squares problem, SIAM Journal of Numerical
Analysis, Vol. 17, p. 883, 1980.
[7] S. V. Huffel, J. Vandewalle, The Total Least Squares
Problem: Computational Aspects and Analysis, 1st Ed.,
SIAM, Philadelphia, 1991.
[8] O. Akyilmaz, Total least squares solution of coordinate
transformation, Survey Review, Vol. 39, p. 68, 2007.
[9] S. Dogan, M. O. Altan, Total least squares registration
of images for change detection, ISPRS Archive Vol.
XXXVIII, Part 4-8-2-W9, Core Spatial Databases
Updating, Maintenance and Services from Theory to
Practice, Haifa, Israel, 2010.
[10] O. O. Ayeni, Statistical adjustment and analysis of
data, A Manual, in the Department of Surveying &
Geoinformatics, Faculty of Engineering, University of
Lagos, Nigeria, 2001.
1993