Automated Vehicle System Using Artificial Neural Network with Mobile Phone driven capabilities

Supervisor: Firoz Mahmud Lecturer, Department of Computer Science & Engineering, Rajshahi University of Engineering & Technology. Author: Al Arafat Department of Computer Science & Engineering, Rajshahi University of Engineering & Technology.

Abstract: This paper illustrates on automated vehicle system using a novel classifier type of artificial neural network, Kohonen-type Concurrent Self Organizing Map (CSOM). This automated vehicle system uses only visually identification of road direction. The path to be identified has been classified into three output directions- left, straight, right for each of which individual SOM modules are used and merges to go through the CSOM classifier module to find the output. This vision based automated vehicle system can also be controlled by mobile device. The operator mobile phone makes a call to the mobile attached to the vehicle. Mean time if any button is pressed, it makes a DTMF tone at the receivers end. A DTMF decoder is used to decode the DTMF tone and a microcontroller is used to make the vehicle know the decision and follow it. One of the DTMF tone is used to turn the mobile phone controlling system off and change the vehicles transmission to automatic state.

1. Introduction: Its a great challenge for the researchers to build a robust autonomous vehicle system still now, although intensive research is going on to make this system more dynamic and more robust. Precise automation of a system lies in two practical principle tasks- sensing and reasoning. More sensors make a system running more precisely by gathering more information about the robot with respect to the surroundings. While the reasoning makes decisions in order to control the robot by applying appropriate algorithms on the information. As human mind always think of innovative & more affordable thinking, mobile controlled robotic vehicle system is an outcome of it. If one wishes not to let his robot on automation he can drive it by his own using his mobile device. This brings a great inspiration in the field of automation of vehicle system.

Artificial Neural Network, a vital part of artificial intelligence which was invented and modeled by being inspired from human brain functionality. As human brain can processes various information to make decision in many variant and uncertain conditions, this specialty of human brain influenced the design & functionality of artificial neural network. Thus an artificial neural network can learn from its given information and adapts to the situation using previous experience. Automated Vehicle System based on traditional vision based pattern recognition may perform well but wouldnt be precise, since path detection in autonomous vehicle system in real time is the main aspect of this system, which is a difficult task. Its because of noise and incessant change in surrounding circumstances in real world scenes. Let us accentuate on the artificial neural network. Artificial neural network proved itself much more promising in doing jobs like pattern recognition, image processing, etc. Artificial neural network has done these dynamic tasks with a higher success rate. Self organizing map, an unsupervised neural network widely used for exploratory data analysis, visualization of multidimensional data, segmentation of complex data etc. Self organizing map is thus used in dynamic pattern recognition in different applications. Autonomous vehicle system can be more robust and user friendly by making it capable of being controlled by mobile phone. DTMF (Dual Tone Multi Frequency), an alternative of RF (Radio Frequency) communication offers a longer distance communication than RF and is now a day widely used to communicate between robot to robot or to device. [4] In this paper we presented an approach to identify visually the road direction of an autonomous vehicle system using self organizing map classifier and to control the autonomous system (mobile robot in our case) alternatively using mobile phone.

2. Approach: The main to tasks of this visual based autonomous vehicle system arei. Input processing, ii. Reasoning a. Feature selection, b. Classification. iii. Feeding output. To control the robot by mobile phone, the main task is to program the microcontroller properly. 2.1 For automated vehicle system using ANN: 2.1.1 Input processing: As this autonomous vehicle will identify road direction visually, it uses camera to take the surrounding images. The images are fed to the processing algorithm as a vector matrix to make decisions.

2.1.2 Reasoning: Reasoning part of this system includes two vital processes of the operating algorithm. The first one performs the feature selection process and the second one includes the classification process.

2.1.2.1 Feature selection: The feature selection method will cover the edge detection of the input images. Edge detection method discards the entire unwanted phenomenon from the image to do more smooth processing over the image. Canny Edge Detector will be used as the edge detection algorithm as it is very popular in detecting optimal edge in an image. [5] 2.1.2.2 Classification: The gen of the system is the classifier, which will make decision for the system. Concurrent Self Organizing Map (CSOM) classifier will be used as the classification model. Classification evolves through two processes. 2.1.2.3.1 Architecture: Three identical SOM modules will be used which will individually direct one direction (left, right or straight). The input vector will be approached to all three modules.

Fig 1: The architecture of the CSOM classifier [1]

2.1.2.3.2 Training and Recognition: As Winner takes all is the working principle of the classifier, the nearest neuron from the input vector will be the winner and will take over the input vector and the module containing the neuron will be assigned to the image (input vector) and the robot will be thus directed. 30 images (10 images for each module) will be used to train the network. Road images from the CMU (Carnegie Mellon University) Vision and Autonomous Systems Center's Image Database will be used. This database includes a large number of road images captured as part of the extensive research they conduct at their Robotics Institute, for the NAVLAB series of vehicles. [3]

Fig 2: Classification of CSOM model [1]

2.2 For mobile controlled vehicle system: In continuing a call, when a button is pressed the corresponding tone is heard at the other end. This tone is defined as DTMF tone. This DTMF tone will be used to control i.e. to direct the robot. The whole process steps likei. The DTMF tone will be received by another mobile phone attested with the robot. ii. The DTMF tone will be decoded by a DTMF decoder and will create binary numbers and sent them to the microcontroller. iii. The microcontroller will decide over the input and direct the robot. The callee mobile phone will act as remote.

Fig 3: Mobile controlled paradigm

3. Preliminary experimental initiatives: This is the basic primary block diagram upon which Im working now, is shown bellow.

Fig 4: General block diagram of the automated vehicle system using CSOM [1]

REFERENCES 1. V.E. Neagoe, M. Valcu, B. Sabac, A neural approach for detection of road direction in autonomous navigation, in Computational Intelligence (Theory and Applications, ed. B. Reusch), Proc. Internat. Conf., 6th Fuzzy Days Dortmund, Germany, May 1999, Springer, Berlin, pp. 324-341. T. Kohonen, Self-Organizing Maps, Springer-Verlag, 1995. D.A. Pomerleau, Efficient training of artificial neural networks for autonomous navigation, Neural Computation, vol. 3, no. 1, 1991, pp. 88-97. Raghavendra, Rahul, Mobile phone operated land rovar www.wikipedia.org

2. 3. 4. 5.