Object Tracking Using Mean Shift and Kalman Filter: A Performance Comparison on Public Datasets

Abstract

Object tracking in video remains a fundamental computer vision challenge, especially when faced with real-world complexities like occlusion and dynamic motion. This review offers a comparative analysis of two enduring methodologies, the Mean Shift algorithm and the Kalman Filter, focusing on research published between 2020 and 2025. Mean Shift, a non-parametric tracker, relies on appearance features, while the Kalman Filter, a state estimator, models object motion. We synthesize recent findings on their design, performance, and limitations, drawing on evaluations from standard benchmarks like MOT, KITTI, and PETS. The analysis highlights a strong trend towards hybrid approaches that leverage the complementary strengths of these classical techniques to achieve robust, realtime tracking in demanding scenarios.