Pedestrian detection based on the improved HOG features

For HOG features' characteristics of high accuracy and large amount of calculation, selected MultiHOG features instead of traditional HOG by means of adjusting the structure of HOG features and using Fisher selection criteria. For the further detection effect, coalesced LBP feature which good at texture based on MultiHOG. The algorithm combining additive cross the SVM classifier to reduce the test time, improved the efficiency of detection and detected pedestrians sliding window. Finally, tested by INRIA standard data sets. The results showed that the algorithm has better feature detection and detection time than traditional one.