Car tracking by quantised input LMS, QX-LMS algorithm in traffic scenes

Abstract

Abstract: The tracking algorithm is an important tool for motion analysis in computer vision. A<br><br>new car tracking algorithm is proposed which is based on a new clipping technique in the field<br><br>of adaptive filter algorithms. The uncertainty and occlusion of vehicles increase the noise in<br><br>vehicle tracking in a traffic scene, so the new clipping technique can control noise in prediction<br><br>of vehicle positions. The authors present a new quantised version of the LMS, namely the<br><br>QX-LMS algorithm, which has a better tracking capability in comparison with the clipped LMS<br><br>(CLMS) and the LMS and also involves less computation. The threshold parameter of the<br><br>QX-LMS algorithm causes controllability and the increase of tracking and convergence properties,<br><br>whereas the CLMS and LMS algorithms do not have these capabilities. The QX-LMS algorithm is<br><br>used for estimation of a noisy chirp signal, for system identification and in car tracking<br><br>applications. Simulation results for noisy chirp signal detection show that this algorithm yields a<br><br>considerable error reduction in comparison to the LMS and CLMS algorithms. The proposed<br><br>algorithm, in tracking some 77 vehicles in different traffic scenes, shows a reduction of the tracking<br><br>error relative to the LMS and CLMS algorithms.