叶语同, 李必军, 付黎明. 智能驾驶中点云目标快速检测与跟踪[J]. 武汉大学学报 ( 信息科学版), 2019, 44(1): 139-144, 152. DOI: 10.13203/j.whugis20170146
引用本文: 叶语同, 李必军, 付黎明. 智能驾驶中点云目标快速检测与跟踪[J]. 武汉大学学报 ( 信息科学版), 2019, 44(1): 139-144, 152. DOI: 10.13203/j.whugis20170146
YE Yutong, LI Bijun, FU Liming. Fast Object Detection and Tracking in Laser Data for Autonomous Driving[J]. Geomatics and Information Science of Wuhan University, 2019, 44(1): 139-144, 152. DOI: 10.13203/j.whugis20170146
Citation: YE Yutong, LI Bijun, FU Liming. Fast Object Detection and Tracking in Laser Data for Autonomous Driving[J]. Geomatics and Information Science of Wuhan University, 2019, 44(1): 139-144, 152. DOI: 10.13203/j.whugis20170146

智能驾驶中点云目标快速检测与跟踪

Fast Object Detection and Tracking in Laser Data for Autonomous Driving

  • 摘要: 利用实时车载激光点云,实现城市环境下的多目标快速检测与跟踪。动态目标跟踪是实现城市环境下自动驾驶的关键,是三维城市场景感知的研究难点。相比于图像,三维激光点云数据更适合用于目标三维形状估计和运动预测,所以广泛应用于无人驾驶方案中。使用基于目标模型和卡尔曼滤波的目标跟踪框架,针对稀疏点云数据中常见的过分割和欠分割问题,提出一种关联历史跟踪结果和目标检测的快速跟踪算法。将跟踪结果作为先验知识,与下一时刻的目标检测关联,增强目标检测的稳定性。该算法已经应用到搭载三维激光扫描仪的自动驾驶汽车中,实验证明,该算法适用于城市交通场景,且满足实时解算需求,单帧处理平均耗时58 ms。

     

    Abstract: A fast algorithm to detecting and tracking multiple objects for an urban driving environment in multi-layer laser data is proposed in this paper. Since situational awareness is crucial for autonomous driving in complicate urban environments and challenging in 3D city perception. Object detection and tracking with cameras or laser has become a popular research topic. Compared with camera, multi-layer laser data is more suitable to estimate 3D model of object and predict motion. So 3D LiDAR is widely used in autonomous driving system. Model-based object tracking framework used in this paper relies on Kalman filter. We extract segmentation in each layer before clustering, which accelerates our detection step. Considering sub-segmentation and super-segmentation happens from time to time in object detection using sparse laser data, we associate the tracking history information with segmentation processing in a fast way. The proposed algorithm in this paper has been applied to the multi-layer laser set up on our autonomous driving vehicle. Experiments demonstrate the applicability and efficiency of our proposed algorithm under urban driving environment. On average, single frame processing takes 58 ms.

     

/

返回文章
返回