A Vehicle-Mounted Radar-Vision System for Precisely Positioning Clustering UAVs

Abstract

The clustering unmanned aerial vehicles (UAVs) positioning is significant for preventing unauthorized clustering UAVs from causing physical and informational damages. However, current positioning systems suffer from limited sensing view and positioning range, which result in poor positioning performance. In order to tackle those issues, a novel vehicle-mounted radar-vision clustering UAVs positioning system is developed, which achieves precise, wide-area, and dynamic-view sensing and positioning of the clustering UAVs. Moreover, a matching-based spatiotemporal fusion framework is established to mitigate cross-modal and cross-view spatiotemporal misalignment by adaptively exploiting the cross-modal and cross-view feature correlations. Furthermore, we propose an attention-based spatiotemporal fusion method that achieves a trinity projective attention with the unique structure and task-oriented format for effective feature matching and precise clustering UAVs positioning. Our method also exploited the modality-oriented cross-modal feature and the UAV-motion-oriented cross-view UAV spatiotemporal motion feature.We demonstrate the advantages of our proposed framework and positioning method in our developed clustering UAVs positioning system in practice. Experimental results confirm that our proposed method outperforms the benchmark methods in terms of the positioning precision, especially under the occlusion scenarios. Moreover, ablation studies confirm the effectiveness of each unit of our method.