People Counting Method Based On Moving Targets Detection

This paper aims at video surveillance in public security and realizes the passen-ger flow counting by using such methods as background model, motion detection, target tracking and pattern recognition. This paper is of profound significance. Practically, it has great commercial value. Technically, it is the cutting-edge research in computer vision domain.Taking no omission and less false positives as the goal, this paper focuses on the study of people counting method with better stability. Firstly, this paper introduces the background and status quo of intelligent surveillance and people counting, and in addition, it presents relevant knowledge of background model, machine learning method and pedestrian features. Secondly, based on running average background model and with mathematical morphology, the effectiveness of motion detection is improved, and then Band-like scanning algorithm which applies to the pedestrian in a long-range is used and its good counting effect is achieved. Based on OpenCV and LibSvm, the feature of HOG is extracted, the counting of detector is computed, and further more, the head-shoulder recog-nition of pedestrian is realized. After the analysis of adaptability of learning samples in different scenes and the study of Re-training, the accuracy of detec-tion is improved remarkably. Finally, this paper envisages the further research of counting algorithm according to current results.