Tracking Multiple Variable-Sizes Moving Objects in LFR Videos Using a Novel genetic Algorithm Approach

All available methods for tracking motion objects in videos have several challenges in many situations yet. For example, Particle filter methods cannot track objects that have variable sizes within frames duration efficiently. In this work, a novel multi-objective co-evolution genetic algorithm approach is developed that can efficiently track the variable size objects in low frame rate videos. We test our method on the famous PETS datasets in 10 categories with different frame rates and different number of motion objects in each scene. Our proposed method is a robust tracker against temporal resolution changes and it has better results in the tracking accuracy (about 10%) and lower false positive rate(about 7.5%) than classic particle filter and GA methods in the videos which contain variable size and small objects. Also it uses only 5 frames in each second instead of 15 or more frames.