Detection and classification of object movement - an application for video surveillance system

Moving and non-moving object detection has never been ignored in the research of computer vision due to its practicality for computer surveillance system. In this paper, the concern is not only detecting and differentiating whether an object is moving or non-moving, but also, classifying grouping objects and speed changing objects as suspicious moving objects. A background subtraction technique has been employed in this work as it is able to provide complete silhouettes of the moving objects. However, it is extremely sensitive to dynamic changes such as change of illumination. The detected foreground pixels usually contain noise. These noises are filtered by a median filter. The classification algorithm makes use of the area of objects to determine whether it is a group or single moving object. Furthermore, the detection of speed changes makes use of the centroid location of each object, and it is calculated by the distance where the centroid has moved across the frames. The proposed system is tested with 30 datasets and the experimental results show that the system has achieved an accuracy of 85%.