基于激光点云与图像信息融合的交通环境车辆检测

随着无人驾驶汽车的快速发展,仅依靠单一传感器的环境感知已经无法满足车辆在复杂交通场景下的目标检测需求。融合多种传感器数据已成为无人驾驶汽车的主流感知方案。提出一种基于激光点云与图像信息融合的交通环境车辆检测方法。首先,利用深度学习方法对激光雷达和摄像头传感器所采集的数据分别进行目标检测;其次,利用匈牙利算法对两种目标检测结果进行实时目标跟踪,进而对两种传感器检测及跟踪结果的特征进行最优匹配;最后,将已匹配及未匹配的目标进行择优输出,作为最终感知结果。利用公开数据集KITTI的部分交通环境跟踪序列及实际道路测试来验证所提方法,结果表明,所提融合方法实时车辆检测精度提升12%以上、误检数减少50%以上。

Vehicle detection in the traffic environment based on the fusion of laser point cloud and image information

With the rapid development of driverless cars, the environment perception solution relying on single sensor cannot meet the demands of vehicles target detection in complex traffic scenarios. Fusion of multiple sensors has become a mainstream perception solution for driverless vehicles. In this study, a vehicle detection method in traffic environment based on the fusion laser point cloud and image information is proposed. Firstly, the deep learning method is used to detect the object data collected by the lidar and the camera sensor. Secondly, the Hungarian algorithm is utilized to track the target detection results in real time. Then, the characteristics of the detection and tracking results from the two sensors are optimally matched with each other. Finally, the matched and unmatched targets are picked and outputted as the final perception results. The proposed algorithm is evaluated in some traffic environment tracking sequence of the public dataset KITTI and real road testing. Experimental results show that the real time vehicle detection accuracy of the proposed fusion method increases more than 12% and the number of false detection decreases more than 50%.