Measuring the Effects of Occlusion on Kernel Based Object Tracking Using Simulated Videos

AbstractOcclusion handling is one of the most studied problems for object tracking in computer vision. Many previous works claimed that occlusion can be handled effectively using Kalman filter, Particle filter and Mean Shift tracking methods. However, these methods were only tested on specific task videos. In order to explore the actual potential of these methods, this paper examined the tracking methods with six simulation videos that consider various occlusion scenarios. Tracking performances are evaluated based on Sequence Frame Detection Accuracy (SFDA). The results show that Mean shift tracker would fail completely when full occlusion occurs as claimed by many previous works. In most cases, Kalman filter and Particle filter tracker achieved SFDA score between 0.3 and 0.4. It demonstrates that Particle filter tracker fails to detect object with arbitrary movement in one of the experiments. The effect of occlusion on each tracker is analysed with Frame Detection Accuracy (FDA) graph