Pedestrian detection based on merged cascade classifier

Reliable pedestrian detection is an important problem in visual surveillance. This paper presents a combined method for pedestrian detection, which significantly improves the detection accuracy without degradation of the detection speed. Firstly, we create a table to mark foreground pixels by means of background difference for eliminating background interference areas. Secondly, Weighted Linear Regression Model embedded into cascade GAB is used for training weak classifier with parts model based on head-shoulder. Finally, two classifiers based on Haar-Like features and Shapelet features respectively fuse to detect pedestrian. The experimental results show that our method can boost the detection rate and reduce the false alarm with non-degradation of detection speed. Particularly, our detection mechanism performs well in the lower resolution and relative complex background situation.