Real-time recognition of moving direction of a target object based on historical trajectories

This paper presents an effective algorithm to identify moving direction of an object in images based on historical trajectories of the moving object. By continuously acquiring images, temporal differencing is performed to detect the moving object followed by median filtering to reduce noises in the images. After that, opening operation is adopted to remove small objects lacking momentum in the binary image and strengthen the surrounding thickness of those with sufficient momentum. Historical trajectories can therefore be constructed by decreasing the momentum of the binary images as time elapsed based on established rules to decide the moving direction of objects.