A Method to Extract Overall Trajectory Information from Frame Sequence of Fixed Background without Object Tracking

In this paper, a method is proposed to extract overall motion pattern for target trajectories from RGB frame sequence in a fixed background. Compared to existing methods which perform tracing on individual targets for trajectories, proposed method extracts overall trajectory information without matching or alignment between surveillance video frames, bringing about reduction on computation cost Rather than tracking every object, researchers detect objects in each frame and record their central positions to form a 2D trajectory points cloud. A novel clustering algorithm based on the moment of inertia has been applied on the 2D point cloud to extract overall statistics trajectory information. In a traffic surveillance scenario, the proposed method provided descriptive information with efficiency, and is suitable for real-time intelligent monitoring and crowd management for future smart cities.