Research And Realization Of Personnel Retrieval And Tracking System In Surveillance Video Image

With the development of computer network communication technology and multimedia technology, we can see video monitoring equipment everywhere in our daily life. Use of the video monitoring equipment, we can make warning before bad events happen, vehicle license plate recognition, the key person retrieval and person behavior recognition. Despite the various research directions in the field has made certain achievements, but there is no mature system able to complete the various functions, to accomplish the task still needs a lot of manual intervention.This paper research the certain people retrieval and tracking through surveillance video. We get real-time video streaming through monitoring equipment for monitoring scene, then make the video to frame, through the detection and tracking the pedestrians in the frame to get pedestrians height, clothes color and human face image, finally we use these message to find the similar person in our current database. We want to build a real-time operational video monitoring system and study the algorithm using in the modules of the surveillance system. The main content of the paper:(1). Pedestrians detection and information extraction(Using current algorithm we can’t real-time detect the pedestrian, we use Gaussian background model and integral image to make pedestrian detection faster; Getting clothes color feature in HSV color space; Calculate pedestrian height through homography matrix and projection matrix)(2). Research suitable people tracking algorithm(Compare famous tracking algorithm compressive tracking and tracking-learning-detection, through compare and analyze we choose the proper algorithm TLD for our real scenes)(3). Face detection and recognition(include face detection through haar-like feature,face alignment using Active Shape Model and our new idea about complete binary tree for face quick search)(4).Two cameras synergy method research (In our real-scene, camera with small focal length has a bigger field, we can get full pedestrian in it,but we can’t acquire clear face witch can be used for extracting feature; camera with big focal length has a smaller field,although we can get clear face,it is easy to track lose. From the above,we need research two camera work together, on the basis of the calibration about projection matrix and homography matrix for cameras,we came up with the hypothesis dual camera detection area matching method based on pedestrian move on the plane)...