Research On Human Detection Technology In Surveillance Video Of Elevated Road

Presently human detection in surveillance video has become to be the focus of research field in intelligent video Surveillance and has great research and application value. Human detection in different application environment is still to be the trend in this field due to the diversity of the application. This paper aims at pedestrian detection on urban elevated road. Considering in such application, face occupies less pixel points in the picture and human may be occluded by other moving objects, so it is difficult to detect the human only using shape information, such as human body shape, eyes, gestures, etc. Therefore, this paper proposes to detect human from face aspect. According to this idea, research work mainly contains two parts. One part focuses on motion region detection between video frames. The second part is on how to detect human face in moving object area. Main research work is bellow:Firstly, systematically summarize and analysis the common used motion targets detection methods. Through comparing their advantages and disadvantages, paper adopts motion targets detection algorithm between two frames in video to extract the motion targets area.Secondly, on the basis of discussion of many face detection methods, paper adopts the face detection method based on skin color. Through many researches done on the cluster characteristic of human face skin color in different color spaces, paper chooses to establish skin model in YCbCr color space and uses this skin model to do skin detection in the motion target areas.Thirdly, considering there may be some moving objects has the similar color with human face skin, such as human bending arms, shaken branches, etc, criteria condition is set to exclude such disruptors by using length to width ratio, candidate region pixels in video image.Finally, real- time pedestrian detection in surveillance video. Detection algorithm is simulated on Matlab7.7 software platform and compared with other detection algorithm. Results show that this method can not only accurately detect human, but also has good real-time performance and little effect of gradual illumination change.