Visual tracking of vehicles using multiresolution analysis and neural network

This paper describes the procedure for detectionand tracking of a vehicle from an on-road image sequence takenby a monocular video capturing device in real time. The mainobjective of such a visual tracking system is to closely followobjects in each frame of a video stream, such that the objectposition as well as other geometric information are alwaysknown. In the tracking system described, the video capturingdevice is also moving. It is a challenge to detect and track amoving vehicle under a constantly changing environment coupledto real time video processing. The system suggested is robust toimplement under different illuminating conditions by using themonocular video capturing device. The vehicle trackingalgorithm is one of the most important modules in anautonomous vehicle system, not only it should be very accuratebut also must have the safety of other vehicles, pedestrians, andthe moving vehicle itself. In order to achieve this an algorithm ofmulti resolution technique based on Haar basis functions wereused for the wavelet transform, where a combination ofclassification was carried out with the multilayer feed forwardneural network. The classification is done in a reduceddimensional space, where Principle Component Analysis (PCA)dimensional reduction technique has been applied to make theclassification process much more efficient. The results show theeffectiveness of the proposed methodolog