The Research Of Pedestrian Detection In Complex Scenes

Pedestrian detection is widely used in vehicle auxiliary driving, communityintelligent surveillance, human-computer interaction, war surveillance etc, which is oneof the research hot spot in the field of computer vision currently.This paper take fixed cameras and mobile cameras pedestrian detection in the singleviewpoint as the research object. On the basis of the background dynamic change incomplex scenes and pedestrian posture changing pedestrian detection and mark. Themain research contents and innovation are as follows:1) On the basis of complex dynamic scenes in fixed camera. Classical Codebookhas large memory requirements, calculated quantity, Codewords created are not exact.This paper proposes object detection in dynamic scenes based on Codebook withsuperpixels(CBSP-OD). CBSP-OD take superpixels segment in HSL color space, anduse L as light value to judge, which overcomes large memory requirements, calculatedquantity. Besides, CBSP-OD is robust to illumination change. CBSP-OD use superpixelsinstead of single pixel to build background model, in which consider spatial consistencywell and create codewords more exactly. In this way, large memory requirements andcalculated quantity can be avoided again. Experiments in the dynamic scenesdemonstrate the proposed method outperform recent state-of-the-art methods. The speedof CBSP-OD is65frame per second, which meet most of real-time systems.2) On the basis of moved camera scene in the single viewpoint, pedestriandetection based on feature classification can not adapt to varied pedestrian size, changedposes, large disturb noise etc, this paper proposed a new algorithm of cascade LBP,SPHOG, SURF adaptively to detect pedestrian(CFA-PD). CFA-PD detail the region ofpedestrian step by step. First, use pedestrian classifier based on LBP to remove largenumber of non-pedestrian area. Second, use pedestrian classifier based on SPHOG narro -w pedestrian area again. At last, use pedestrian classifier based on SURF to detect andmark pedestrian. Experiments demonstrate the proposed method solve the problem ofvaried pedestrian size, changed poses, large noise. Also, CFA-PD detect pedestrianexactly.3) On the basis of varied poses, charged pedestrian size cause pedestrian markwindow not exactly, this paper propose a new algorithm of pedestrian mark windowadapt to the size of pedestrian(PM-OTSU). PM-OTSU take pedestrian window as initialbrief. Change the size of mark window according to the number of initial brief elements.The result of experiments demonstrate PM-OTSU solve the problem of pedestriandetection window automatic update, and very exactly.