The Research Of Human Detection Algorithm In Surveillance Video

Intelligent video surveillance system is an important research direction in the field of computer vision. Intelligent video surveillance system has been widely applied in the field of public safety and human-computer interaction. It is also one of the key technologies of vehicle auxiliary system at the same time. The pedestrian detection is a hot issue in intelligent video surveillance system. Because people is a kind of no rigid objects, the pedestrian appearance is different and the detection environment such as illumination is complex that brings great challenges to the pedestrian detection. The main pedestrian detection methods are based on machine learning today. This paper combines the main pedestrian detection methods with the moving target detection algorithms and the pedestrian tracking algorithms in order to design a pedestrian detection system which has good performance at the accuracy and real-time capability. For the above system, this paper completes the following work:First, this paper studies a variety of background modeling algorithms. We select ViBe model to get the binary mask graph of the moving target because of its accuracy and real-time capability. Then we get a complete mass by using the federated filtering algorithm and obtain basic information of the mass by using three scans connected component labeling algorithm. We also improve the speed and effect of moving target detection module by using image sampling and image denoising.Then, we classify the pedestrian based on machine learning methods and combine the HOG feature with softcascade classifier to do multi-scale detection of pedestrians. We combine the head shoulder of pedestrians with the body of pedestrians to detect people. We also solve pedestrian occlusion problem by using the detection of the head shoulder of pedestrians and reduce the rate of false detection of pedestrians by using the detection of the body of pedestrians. At the same time, we use pyramid search to solve the problem of different scales of pedestrians in the surveillance video.Finally, we use a pedestrian tracking algorithm based on the multi feature template matching in order to improve the speed of the system. The pedestrian tracking algorithm blends the color feature, gradient feature and texture feature of a pedestrian. We also use the movement prediction way to accelerate the search speed of the process of template matching.In summary, the pedestrian detection system consists of a moving target detection module, a pedestrian detection module and a pedestrian tracking module. The whole system performs well in real time capability and detection accuracy.