Research on target tracking methods for Beidou positioning safety helmet communication terminal systems based on an improved ShuffleNetV2 network

To enhance target tracking accuracy and positioning robustness of helmet terminal systems in construction site scenarios, a BeiDou visual fusion method based on an improved ShuffleNetV2 network was developed. This study proposes a multi-scale perception structure optimization strategy, incorporating channel attention mechanisms and mutual information supervision loss to strengthen feature expression capabilities and spatial alignment effects. A Kalman prediction model integrating visual and BeiDou data was designed to achieve dynamic estimation and spatiotemporal coordination of target states. Evaluated using the BD-HelmetTrack dataset, results demonstrate that this method significantly outperforms comparison models under occlusion and multi-target interference conditions, improving IDF1 metrics and tracking stability while maintaining low latency.