Video Surveillance Based On Human Detection And Tracking

Computer Vision is a branch of the Artificial Intelligence, it has been developing rapid recent years companied with the fast developing of the capability of the computer. It is an science that let the computer see the scenes and then make it recognize the objects in the scenes.Event Detection is a quite important field in the Computer Vision. It bases on Object Detection and Tracking, combined with other factors to detect the predefined events in the video, such as cars traveling up against, pedestrian running at the red light and some special events..Nearly every event detection method consists of Foreground Detection and Tracking, Background Refreshing, Extracting of the Foreground Features and the classification by the Classifier. At present there are some methods in event detection such as:optical flow, background subtraction, camshaft, HOG features, Haar Features, Classifier Machine and so on. Each method has its field of good competence, it may become worse in other fields and environment. So only using one method may gets a unpleasant result. In many cases, Integrating two or more methods together to get a better result is necessary.Event Detection is a rapid developing field in the Computer Vision. Though some goals have been achieved such as vehicle license plate recognition, some basic theories are needed which have important in this subject but have not been established. It still truly has good prospects in many application fields, human detection, face recognition have wild application in many situation.This article introduces the event "Elevator NoEntry" detection in TRECVID 2009. Its aim is to detect there is nobody goes into the elevator when the elevator is open until it close. I use the Optical Flow method between two frames to judge whether there is movement in the Video, Use the Hough Transform in the binary image of the foreground to get the status of the elevators. And then find the contours in the foreground to detect out the people, tracking is implanted with camshift., With the integrated methods, the system produces a pleasant performance.