Modified particle filter for object tracking in low frame rate video

Object tracking algorithm using modified Particle filter in low frame rate (LFR) video is proposed in this paper, which the object moving significantly and randomly between consecutive frames in the low frame rate situation. Traditionally, Particle filtering use motion transitions to model the movement of the target. However, in object tracking with low frame rate sequences, it is very difficult to model significant random jumps of subjects. The key notion of our solution is that using the object detection and extraction to locate the tracked object, while not using the dynamical function. We propagate the sample set around the detected regions, which the samples are assumed to be uniformly distributed in the neighborhoods of the detected region. It is similar to the general particle filter to propagate samples. Then we compute the likelihood between the target model and the candidate regions, which are based on color histogram distances. Our extensive experiments show that the proposed algorithm performs robustly in a large variety of tracking scenarios.