基于毫米波雷达和视觉的目标检测方法

Target Detection Method Based on Millimeter-wave Radar and Vision

为了提高目标检测网络对远距离目标的检测能力,以及改善由单一视觉传感器的感知系统抗环境干扰能力差的问题,提出了一种基于毫米波雷达和视觉传感器的多源目标检测方法,视觉图像经由多个毫米波雷达获取的点云信息增强后进行检测。首先,对增强同一视觉图像的多个雷达点云进行数据拼接,通过坐标转换将雷达点云投影至视觉图像平面,并对超出雷达探测距离的异常点和经过坐标转换后位于视野外部的无效点进行剔除,生成雷达点云图像。然后,根据雷达点云图像中各雷达点的位置与深度信息,形成对应的感兴趣区域,生成雷达特征图像。最后,将雷达特征图像与视觉图像在YOLOv4网络中的主干特征提取部分进行多级融合,并使用通道注意力机制分配通道权重。实验结果表明,基于雷达增强的目标检测网络的平均检测精度提高了10.93%,并提高了远距离目标和弱光照条件下的检测精度,具有比传统机器视觉的目标检测方法更好的可靠性和鲁棒性。

In order to improve the ability of target detection network to detect long-range targets and to improve the poor anti-environmental interference ability of the perception system by a single visual sensor,a multi-source target detection method based on millimeter-wave radar and visual sensor is proposed.The visual image is detected after the point cloud information obtained by multiple millimeter-wave radar is enhanced.First,several radar point clouds that enhance the same visual image are stitched,projected to the plane of the visual image through coordinate transformation,and the abnormal points beyond the radar detection distance and invalid points outside the field of view are eliminated to generate the radar point cloud image.Then,according to the position and depth information of each radar point in the radar point cloud image,the region of interest is formed,and the radar feature image is generated.Finally,the backbone of YOLOv4 network is fused with the radar feature image and the visual image,and the channel weight is allocated using the channel attention mechanism.The experiment shows that the average detection accuracy of target detection network based on radar enhancement is improved by 10.93%,and the detection accuracy of long-range target and weak illumination condition is improved.It has better reliability and robustness than traditional machine vision target detection method.