STUDY AND INVESTIGATE THE CUCKOO OPTIMIZATION TO IMPROVE THE PARTICLE FILTER AS A BASIC TRANSITION MODEL ON SCALE VARIATION AND OCCLUSION OBJECT TRACKING

Scale variation and occlusion are still the main problems in visual object tracking. These problems arise because of uncertainty and the unpredictable movement of other objects around the target. A probabilistic-based tracker commonly deals with the problem. However, it is challenging to develop a robust transition model with a high-quality observation model. The proposed method is designing a cuckoo search to optimize Particle Filter as a base transition model. Cuckoo search spreading more various candidate target tracking based on Lévy Flights. The proposed method combined with affine transformation as particle representation and deep learning as an observation model. The proposed method achieves a precision of 0.894 and a success rate of 0.701 on the scale variation problem. It obtains a precision of 0.824 and a 0.621 success rate on occlusion, which is better than the baseline particle filters-based method. It also obtains competitive compared to the state-of-the-art method with seven times faster in computation. Robust Tracker in occlusion and scale variation becomes the fundamental base to real applications such as surveillance, robotics, and other intelligence systems.