雷达和视频融合技术下小目标车辆检测方法

Small target vehicle detection method based on radar and video fusion technology

摘要:

针对复杂道路交通场景下小目标车辆检测难题,提出基于雷达和视频融合的检测方法。将K-means聚类算法与YOLOv5网络损失函数深度融合,构建联合识别框架,通过雷达点云聚类提取目标空间位置信息,利用YOLOv5网络提取视觉特征,实现信息互补,并设计联合损失函数平衡两种模态贡献度。针对空间坐标系转换问题,提出高精度转换方法,结合高斯平面直角坐标系和椭球基准,实现目标位置信息的精确转换,并引入误差校正机制提高准确性。通过小目标数据采集、增强及特征重构,提升数据描述能力与模型鲁棒性。定义动态跟踪标准,结合车辆运动状态和位置信息,实现对小目标车辆的持续识别与定位。实验结果表明,该方法结合雷达与视频技术优势,有效提高了复杂道路交通环境下小目标车辆的检测精度。

Abstract:

A detection method based on radar and video fusion is proposed to address the challenge of detecting small target vehicles in complex road traffic scenarios. Deeply integrate the K-means clustering algorithm with the YOLOv5 network loss function to construct a joint recognition framework. Extract target spatial position information through radar point cloud clustering, use the YOLOv5 network to extract visual features, achieve information complementarity, and design a joint loss function to balance the contributions of the two modalities. Aiming at the problem of spatial coordinate system conversion, a high-precision conversion method is proposed, which combines Gaussian plane Cartesian coordinate system and ellipsoid reference to achieve accurate conversion of target position information, and introduces error correction mechanism to improve accuracy. By collecting, enhancing, and reconstructing small target data, the data description ability and model robustness are improved. Define dynamic tracking standards, combined with vehicle motion status and position information, to achieve continuous recognition and localization of small target vehicles. The experimental results show that this method combines the advantages of radar and video technology, effectively improving the detection accuracy of small target vehicles in complex road traffic environments.