Visual Target Locking during Fast Ground Maneuver using Enhanced ORB Predictive Particle Filter

Maintaining the visual lock on a target during rapid ground maneuvers is critical for various applications, including autonomous vehicles, robotics, and defense systems. Existing visual tracking algorithms face challenges in maintaining accurate and reliable target locks in dynamic and unpredictable maneuvers, especially on previously unknown targets. One technique developed to address this issue is the ORB-enhanced particle filter (ORBPF). However, ORBPF is unable to handle quick maneuvers because the prediction method uses a constant velocity (CV) dynamic model. This research focuses on addressing the limitations of existing ORBPF tracking algorithms for fast ground maneuver objects using the constant-velocity and constant-turn (CVCT) dynamic model approach to predict the next movement. By incorporating motion prediction into the current ORBPF framework, the proposed system anticipates future target positions and adaptively adjusts tracking parameters to maintain stable locking. Functional testing was performed using an object tracking benchmark to compare ORBPF-CVCT with standard ORBPF in specific scenarios. Results show that the standard ORBPF-CV achieved RMSE up to 345,518 px while our proposed ORBPF-CVCT achieved up to 191,387 px. These results show that the proposed method is able to track fast maneuver objects better than the previous ORBPF method.