重庆理工大学学报2025，Vol.39Issue(5)：43-50,8.DOI:10.3969/j.issn.1674-8425(z).2025.03.006
基于激光雷达的FSAC赛道环境感知研究
Research on environment perception of FSAC racetrack based on LiDAR
摘要
Abstract
To address the reduced accuracy in detecting cone bucket in the sensing system of driverless formula racing cars,we propose a 3D object detection method of cone bucket on FSAC track,which combines artificial feature extraction and deep learning.First,the original point cloud is extracted with areas of interest,voxelized grating filtering,ground plane filtering and European clustering to generate a target detection frame of the first stage.Then,the original point cloud data is represented based on cylindrical elements and features are extracted point by point using PointNet idea.Each point is restored to H and W resolution to form a pseudo-image.The convolutional neural network is employed to process the pseudo-image results and the SSD target detection network is introduced to generate a two-stage target detection frame.The two-stage detection frame is calculated through the detection frame confirmation module.The intersection ratio of the two-stage detection frame is considered as a cone bucket if the intersection ratio is greater than the set threshold.Finally,the experiment is conducted on campus by imitating the environment of the competition field.Our experimental results show the cone bucket target detection algorithm combined with artificial feature extraction and deep learning delivers a fairly good performance.The missed detection rate of the algorithm is 0.08%,the false detection rate is 0.02%,and the frame rate is about 60 FPS,improving the racing cars'environmental perception capability.关键词
车辆工程/无人驾驶方程式赛车/激光雷达/目标检测/深度学习Key words
vehicle engineering/driverless formula racing car/LiDAR/object detection/deep learning分类
交通运输引用本文复制引用
佟战龙,李刚..基于激光雷达的FSAC赛道环境感知研究[J].重庆理工大学学报,2025,39(5):43-50,8.基金项目
辽宁省自然科学基金资助计划项目(2022-MS-376) （2022-MS-376）
辽宁省教育厅重点攻关项目(JYTZD2023081) （JYTZD2023081）
