Real-Time Occlusion-Aware Object Tracking

This paper presents a novel system for real-time object tracking in occlusion-prone environments, addressing critical challenges in achieving robustness, accuracy, and efficiency under dynamic and cluttered conditions. The proposed system integrates geometric analysis using Intersection over Union (IoU), probabilistic validation via a Hidden Markov Model (HMM), and an advanced object detection module to ensure consistent performance during both partial and complete occlusions. Evaluated on the KITTI dataset, the system achieves a Success Rate of $\mathbf{8 3 \%}$, Precision of $\mathbf{9 3 \%}$, Recall of $\mathbf{8 6 \%}$, and an F1 Score of 0.91, outperforming state-of-the-art methods such as Query-Guided Redetection Tracker (QRDT) and Siamese Occlusion-Aware Network (SiamON) in terms of accuracy and reaction time, with a low processing delay of 35 ms. While QRDT and SiamON offer competitive results, the proposed system demonstrates superior resilience to prolonged occlusions and enhanced suitability for time-critical applications like autonomous driving and surveillance. Limitations related to identity consistency during extended occlusions are acknowledged, with future directions focused on incorporating advanced trajectory prediction and multi-modal sensor fusion. This research establishes a robust foundation for occlusion-aware tracking and contributes significantly to the domains of autonomous systems, robotics, and computer vision