Video Event Backtracking System Based On Face Under Multi-camera

In current society,video surveillance system has become an important means of case detection and the suspicious target person confirmation,but manual search is easy to miss some useful information with time-consming,low efficiency.For the problem of low efficient,the paper developed a video event backtracking system based on face under multi-camera in the park environment.Firstly,face detection for all the camera's video in the system,obtaining all of the faces that appeared in the cameras;secondly,faces tracking on the camera's videos using inter-frame constraint algorithm,obtain multiple faces of each person as face database,and then Face recognition on the object face that set by box in the videos with the face database;Finally,to obtain the time and the walking path of the target person by the result of face recognition,and get the backtracking result of the object.The main contents of this article include:Adaboost face detection algorithm has the feature of higher detection rate and real-time ability,so the method is used in face detection.However,there are part of false positive in the presence of test results,for this problem the paper use skin color detection to reduce the false positive rate.And further use inter-frame constraint improves detection rates.This method determine the large probability region of the face in the next frame based on the previous frame test result,then weaken classifier in this region and reduces the detection conditions,in order to improving the detection rate.Because of the face detection is needed for each camera,so a combination of inter-constrained which on the basis of face detection is used to conduct face tracking.Firstly,determine the large probability region of the face in the next frame based on the previous frame test result,and then regard the detected face in the region as a tracking result.This method makes use of the video features,make single sample face database become multiple sample face database,easy to imply,which will in favor of face detection.Face Recognition use block LBP feature,which is a good description of the local texture information and illumination robustness.Then sorted the result according the distance of feature for recognition then confirm object face.These can increase the recognition accuracy.The result showed that this system,intelligent method which is the time that target pedestrians appear in video camera is obtained,it can efficiently realize the goal of back tracking.This method can quickly find its whereabouts when the events occur,without artificial viewing the video,in order to reduce the workload of manual searching.