基于DST融合多视图模糊推理赋值的三维目标检测

针对前置激光雷达的点云数据,提出一种基于DST融合多视图模糊推理赋值的有效障碍物分割判别方法.将点云数据转换为体素地图并进行路面分割,得到前、俯视图.在两视图中根据不同的模糊推理规则对某体素属于目标的程度进行基本概率赋值,并通过DST融合判别目标,精确分割目标,从而得到方盒模型参数.将三维识别问题转换为一系列的二维检测问题,与直接利用三维点云信息相比,可以降低数据处理复杂度,提高系统稳定性.在自主研发的自动驾驶汽车上采用前置16线激光雷达和TX2嵌入式开发板进行多次在线试验,并在KITTI上进行对比验证,结果表明所提方法在实际应用中拥有较好的实时性和准确性.

3D object detection based on DST fusion multi-view fuzzy reasoning assignment

An effective object segmentation and discrimination method based on DST fusion multi-view fuzzy reasoning assignment is proposed for the point cloud data of pre-lidar. The point cloud data is transformed into a voxel map and road surface segmentation is carried out to obtain the front and top views. In the two views, the basic probability of the voxel''s degree of belonging to the object is assigned according to different fuzzy inference rules, and the target is distinguished by DST fusing, and the object is accurately segmented to obtain the parameters of the box model. Converting the 3D recognition problem into a series of 2D detection problems, compared with directly utilizing the 3D point cloud information, can reduce the complexity of data processing and improve the stability of the system. Experiments have been carried out on a self-developed autonomous cars using a 16-line lidar and TX2 embedded development board, with the comparison and verification on KITTI. The results show that the method has good real-time and accuracy in practical application.