视觉与激光雷达信息融合的目标领航车识别方法

摘要

In order to improve the accuracy and real-time of vehicle detection and recognition in unmanned ground vehicle (UGV)Leader-follower Systems,a detection and recognition method of the pilot vehicle based on LIDAR and visual camera information fusion is proposed. Before the recognition of the vehicle system,it is necessary to calibration two kinds of sensors. Determining the relationship between the three coordinates of the LIDAR coordinates,the camera coordinates and the vehicle body coordinates. In the process of vehicle identification,the nearest neighbor clustering analysis is performed on the LIDAR data,and the target is selected based on the clustering result. According to the conversion relation between the LIDAR data and the camera image,Determine the region of interest of the hypothetical target in the image. Then the symmetry feature and corner features of the region of interest are extracted,and the multi-feature cascade classification method is used to identify whether the hypothetical target is the target pilot vehicle. Finally,according to the result of verification of the ROI,it is fed back into the LIDAR data to locate the distance and angle of the pilot vehicle with respect to the following vehicle. The experimental results show that the proposed algorithm has good adaptability to the environment,and makes up for the deficiency of single sensor in vehicle detection and recognition.%为提高军用地面无人平台自主跟随过程中对目标领航车辆检测识别的准确性和实时性,提出了一种基于激光雷达和视觉摄像机信息融合的目标领航车检测识别方法.在识别系统工作前,需对激光雷达和摄像机两种传感器进行标定,确定激光雷达坐标、摄像机坐标、车体坐标三者之间的相互关系.车辆识别过程中,首先对激光雷达数据进行最近邻域法聚类分析,根据聚类的结果对周围目标进行初步筛选生成假设目标;根据激光雷达数据和摄像机图像之间的转换关系,确定假设目标在图像中的感兴趣区域,可以有效减少图像处理的计算量;最后,利用多特征级联分类识别方法验证假设目标是否为目标领航车辆.实验结果表明该算法具有较好的环境适用性,弥补了单目视觉传感器在目标领航车辆检测识别过程中无法检测到深度信息,以及激光雷达不能准确判断出目标为何物的不足.