复杂环境下雷达-视觉目标融合技术

Radar-Visual Target Fusion in Complex Environment

摘要:

针对以往复合探测系统目标跟踪效果不理想、环境适应性不强等问题,提出了一种基于航迹关联融合的雷达-视觉融合系统。利用深度学习、视频结构化技术及航迹聚类跟踪算法生成视觉、雷达目标航迹,并进行时空同步、边界框聚类等预处理;基于修正k-最近邻(modified k-nearest neighbor, MK-NN)法和三级关联约束条件进行航迹关联融合处理,并引入定位几何精度衰减因子(geometric dilution of precision, GDOP)作为融合系数,充分利用雷达、视觉局部航迹得到准确完整的目标航迹信息。在多目标场景下进行试验验证,结果表明该融合系统能够实时稳定跟踪目标,目标编号(ID)跳变率为3.45%,系统航迹符合率达97.6%,目标跟踪的准确率、稳定性得到大幅提升。系统适用于交通监控、周界防护等复杂场景,研究成果也可为反无人机群等军事场景应用提供参考。

Abstract:

Aiming at the problems of poor target tracking effect and weak environmentaladaptability of the previous composite detection system, a radar-vision fusion system basedon track association fusion was proposed. The vision and radar target tracks were generatedseparately by using video structure technology based on depth learning and track clusteringtracking algorithm, and then preprocessed by spatio-temporal synchronization, boundarybox clustering, etc. The track association fusion processing was carried out based onmodified k-nearest neighbor (MK-NN) method and three-level association constraints, andthe geometric dilution of precision (GDOP) was introduced as the fusion coefficient to make full use of radar and visual local track to obtain accurate and complete target trackinformation. The results of the verification in the multi-target scene show that the fusionsystem can track the target stably in real time, the target tracking identify document (ID)jump rate is 3.45%, and the system track coincidence rate is 97.6%, indicating that theaccuracy and stability of target tracking have been greatly improved. The scheme can beapplied to complex scenarios such as traffic monitoring and perimeter protection, and theresearch results can also provide reference for military scenarios such as anti-UAV groups.