This paper presents a study on object tracking in surveillance systems using Particle Filter and Aggregate Channel Features (ACF) detection to address the challenges of accurately tracking multiple objects in dynamic environments. Object tracking is a crucial component in computer vision, with applications spanning from surveillance and security to autonomous navigation and robotics. In this work, we leverage Particle Filter, a robust Bayesian-based filtering algorithm known for its effectiveness in non-linear and non-Gaussian conditions, to track objects consistently over time. The ACF detection method is employed for its high precision in identifying objects across various frames, thereby enhancing initial detection accuracy. Performance testing is conducted across four datasets, using key metrics such as precision, Multiple Object Tracking Precision (MOTP), and Multiple Object Tracking Accuracy (MOTA) to evaluate effectiveness. The results show that while Particle Filter combined with ACF detection achieves consistently high precision (95-99%) and stable MOTP rates (69-79%), challenges arise in maintaining uninterrupted tracking accuracy, as evidenced by lower MOTA scores (3.1-7.2%) and a significant rate of false negatives, especially in complex scenarios with occlusions. These findings suggest that although Particle Filter and ACF detection are effective for initial detection and data handling, enhancements or hybrid methods may be required for applications demanding high accuracy in continuous multi-object tracking.
Keywords—computer vision, multiple-object tracking, particle filter, clear mot, acf detection