Statistics On Human Targets Basing On Motion Detection And Tracking

In the building environment, the number of persons and their staying time are two effective indicators in evaluation of landscape, which reflect the people's fondness of a certain scene. To achieve this goal, we compute those two factors basing on motion detection and tracking techniques on a sequence of images. Our main focus includes background modeling and updating, detection of moving objects and matching track.In this paper, we adopt a statistical approach combined with the average value method to model and update the background so as to adapt to the slow changes of light in the scene. Besides, by dividing the background into blocks, we can accelerate the updating speed for the stationary objects in scene and the'be about to moving'objects. The human body target is obtained by background subtraction on three different methods, which are partition on grey in RGB space, on chroma in HSI space and on similarity of color vectors. All of those three methods perform well. We give a thorough treatment on the connected component of human target and marking scheme, therefore acquiring the basic characteristics of human body.In order to compute the staying time quickly, this paper proposes a rapid backtrack method for locating the key frames to identify the entering time and leaving time. We also give a detailed discussion on the feature extraction and matching, including single target and multiple targets separately.The experimental result shows that our method is fast and accurate. The two factors can offer a useful reference to the evaluation of building environments and scenes.