Algorithm of Head Detection and Tracking Based on Adaboost and Improved Resampling for Particle Filter

摘要

Aiming at non-rigid structure and randomness of pedestrians, we adopt the algorithm of SVM(Support Vector Machine) to extract the HOG (Histograms of Oriented Gradient) features, detect body targets, at the same time, we also adopt the algorithm of AdaBoost to extract the MB-LBP (Multiscale Block Local Binary Pattern) features, detect head targets. Careful contrast of two detection results is remarkable, therefore, we come to the conclusion that the method of detecting head targets is more accurate when there is shelter between pedestrian targets. In process of tracking targets, we improve the original resampling for particle filter algorithm. The experiments show that the state estimation of the improved algorithm of resampling is closer to the true state than that of the original resampling algorithm which can effectively reduce the error of state estimation and the running time.