摘要
针对摄像头在无人驾驶系统车辆检测中易受环境干扰的问题,通过激光雷达数据和摄像头图像进行融合,提出了一种强鲁棒性实时车辆检测算法。首先,将三维激光雷达点云通过深度补全方法转换为和图像具有相同分辨率的二维密集深度图。然后将彩色图像和密集深度图分别通过YOLOv3实时目标检测框架得到各自的车辆检测信息。最后,提出了决策级融合方法将两者的检测结果进行融合,得到了最终的车辆检测结果。在KITTI数据集上对算法进行评估,实验结果表明该算法完全满足无人驾驶车辆所需的强鲁棒性、强实时性和高检测精度的要求。
Aiming at the problem that the camera is vulnerable to environmental interference in the vehicle detection of autonomous vehicle system, a robust real-time vehicle detection algorithm based on LIDAR and camera fusion is was proposed. Firstly, The the three-dimensional LIDAR point cloud is was transformed into a two-dimensional dense depth map with the same image resolution by the depth completion method. Then, the color image and the dense depth map are were respectively used to get the vehicle detection information through the YOLOv3 real-time object detection framework. Finally, a decision level fusion method is was proposed to fuse the two detection results, and the final vehicle detection results are were obtained. KITTI dataset is was used to evaluate the algorithm. The experimental results show that the proposed algorithm fully meets the requirements of strong robustness, strong real-time performance and high detection accuracy required by the autonomous vehicle.
作者
陈毅
张帅
汪贵平
CHEN Yi;ZHANG Shuai;WANG Guiping(School of Electronics and Control Engineering,Chang'an University,Xi'an 710064,China)
出处
《机械与电子》
2020年第1期52-56,共5页
Machinery & Electronics
作者简介
陈毅(1996—),男,陕西西安人,硕士研究生,主要研究方向为无人驾驶汽车环境感知;张帅(1996—),男,陕西延安人,硕士研究生,主要研究方向为移动机器人;汪贵平(1963—),男,湖北麻城人,博士,教授,主要研究方向为智能网联汽车。
