Research Of Moving Human Detection In Video Sequences

The problems of human detection based on video sequence are a hot issue in thecomputer vision and pattern recognition filed. It has extensive application prospect inintelligent monitoring, intelligent vehicle system, graphics and image processing area.Based on the study of domestic and foreign relevant research result and the latestresearch progress, this paper constructed a video-based moving human detection systemfor the problem of moving pedestrian detection in video.Firstly, for video sequences, we modeling background using the mixed gaussianmodel, and detect the motion area based on this model of background. Then a kind ofsegmentation method was proposed for the foreground area of movement which wasdetected by the background model. This method dealt with the foreground movementarea combining median filtering with mathematical morphology processing, and thenexploited projection process segmenting the foreground area of movement. Theexperimental results show that the segmentation method of the foreground movementarea can effectively extract the moving foreground area in video.Secondly, for the pedestrian detection problem, the construction method was deeplystudied for the histograms of oriented gradients characteristic. Then we integrate thecalculation method of integral figure and the construction method of the histograms oforiented gradients characteristic to speed up the computing speed. This algorithmexploited linear support vector machine to train a binary classifier offline, and combinemulti-scale detection and window fusion method based on kernel density estimation todetect and localize the pedestrian in image. Experimental results demonstrate thedetection method can fairly detect and localize the pedestrian appearing in image.Finally, a video-based moving human detection system was constructed bycombining moving object detection with pedestrian detection. Then the video-basedmoving human detection system was realized in engineering by the computer visionlibrary. The experimental results show that this system can detect the moving pedestrianexisting in video effectively.