融合4D毫米波雷达与深度视觉的目标检测与跟踪

Target Detection and Tracking with the Fusion of the 4D Millimeter-Wave Radar and Depth Vision

为解决现有毫米波雷达和传统机器视觉融合方案在复杂环境下车辆检测准确率较低的问题,首先,使用4D毫米波雷达替代传统毫米波雷达,使用自适应卡尔曼滤波算法滤除雷达杂波并跟踪目标;其次,使用车辆数据集训练改进深度视觉网络MobileNetV2+SSDLite来提高视觉识别车辆的准确率;最后,采用决策融合方案完成毫米波雷达信号和视觉信号融合。通过对比不同环境下的实验结果表明,改进方案可以完成对目标车辆的有效估计与跟踪,在不同环境下都有着很好的车辆识别效果,以及更好的可靠性和鲁棒性。

In order to solve the problem of the low vehicle detection accuracy of the existing fusion scheme of the millimeter-wave radar and traditional machine vision in the complex environment,this article firstly uses the 4D millimeter-wave radar to replace the traditional millimeter-wave radar,and uses the adaptive Kalman filtering algo⁃rithm to filter out radar clutter and track targets.Then,it uses vehicle datasets to train and improve the depth vision network MobileNetV2+SSDLite to improve the accuracy of the visual recognition of vehicles.Finally,it uses a de⁃cision fusion scheme to complete the fusion of millimeter-wave radar signals and visual signals.By comparing ex⁃perimental results in different environments,it is shown that the improved scheme can effectively estimate and track the target vehicle,which has good vehicle recognition effects in different environments and has better reliability and robustness.