Automatic Tracking and Human Identification Based on Combination Between YOLO and KCF Algorithm

This research introduces a hybrid tracking system that integrates the YOLO (You Only Look Once) and KCF (Kernelized Correlation Filter) algorithms for enhanced automatic tracking and human identification. The study, conducted at Beihang University, addresses the limitations of the KCF algorithm in handling device offset and target loss by utilizing YOLO for initialization and re-calibration. The Offset Error Rate (OER) serves as a metric to decide when YOLO should update the KCF tracking frame. The proposed method demonstrates improved tracking accuracy and robustness, with potential applications in autonomous navigation and obstacle avoidance for unmanned vehicles. Future work will focus on optimizing the YOLO structure and KCF algorithm for higher efficiency. The research is supported by various foundations, including the Chinese National Natural Science Foundation.