The Study Of Target Face Tracking Based On Surveillance Video

Computer vision technology has made great progress after decades of development from the 1970 s to the present.Face tracking based on surveillance video has become a hot topic in current research.In television and telephone conferences,intelligent monitoring,human-computer interaction,smart city,safe city and other applications,the figure of the face tracking technology can be seen.But face tracking technology based on surveillance video is still facing many challenges in the current practical application,two mains of them are as the following: the first is the problem of the video shooting environment,because most of the videos are filmed under uncontrolled natural conditions and different situations,different lighting and different background conditions occur in different videos and even different frames of the same video;the second issue is the different poses of the same tracking target in one video,a tracking target's exercise in a video is random during video shooting.It does not always face the camera frontally,and it is often partially or completely blocked by obstacles.These are problems that are not easy to be solved in the research of face recognition and tracking.Designing a face tracking technology with high efficiency,good robustness,and strong real-time performance is a common goal pursued by most researchers.Before the target human face in the video is tracked,the target face needs to be detected,therefore,this paper conducts in-depth research on the problem of target face tracking in surveillance video from face detection and face tracking on the basis of previous research results,and achieved the following results:1.In the aspect of face detection,as the traditional adaboost algorithm needs to traverse all the sub-windows in the image to be detected to cause the problem of slow detection speed,this paper proposes to combine the skin color segmentation with the improved Adaboost algorithm to solve it.in the algorithm,skin-color separation in YCgCr and YCgCb spaces is firstly performed to Separate the skin and non-skin areas in each video frame effectively,then morphological operations are performed on the reserved skin tone areas,in addition,human face geometry features are introduced into the skin-color regions obtained by morphological processing to further limit the suspected face regions,eliminate interference from non-human face regions,and reduce the search area of the Adaboost algorithm as much as possible,thereby improving face detection speed.Experimental results show that the face detection speed and the detection effect of the improved Adaboost algorithm have been significantly improved.2.In the aspect of face tracking,as the traditional Camshift face trackings' results are easily interfered with the non-face skin color areas of the human body and the skin-color like backgrounds in the video frame image,this paper presents an improved Camshift face tracking algorithm based on skin-color segmentation,in the algorithm,skin-color segmentation and improved adaboost algorithms are combined with the traditional Camshift to track the target face.The experimental results show that the tracking results of this method are more reliable and the tracking's real-time is also better.