针对自动驾驶车辆在城市开放道路场景下,交通元素和交通事件感知精度不足、目标检测模型参数大且难部署的问题,提出了 一种多视角的轻量化车路协同感知模型.在车辆驾驶过程中分别对路侧和车辆的传感器各自的图像数据进行采集,通过车辆姿态信息与相对位置,将车辆信息转换到路侧坐标系下,实现坐标系的统一;利用Im-VoxelN et算法进行目标检测,同时采用训练后量化(Post Training Quantization,PTQ)方法压缩目标检测模型;通过匈牙利算法对感知结果进行融合,实现城市开放道路场景下高精度、低开销的车路协同目标感知;采用车路协同的数据集进行实验.结果表明该模型实现了城市开放道路场景下车路协同感知的检测能力,同时验证了所提算法的可行性和有效性.
Lightweight Vehicle Road Collaborative Perception Model Based on Multiple Perspectives
A multiple perspective lightweight vehicle road collaborative perception model is proposed to address the issues of insuf-ficient perception accuracy of traffic elements and traffic events,large and difficult deployment of target detection models for autonomous vehicles in urban open road scenarios.Firstly,during driving process of the vehicle,image data of roadside and vehicle sensors are col-lected separately.Through vehicle attitude information and relative position,the vehicle information is transferred to roadside coordinate system to achieve unity of the coordinate system.Then,ImVoxelNet algorithm is used for object detection,and Post Training Quantifica-tion(PTQ)quantization method is used to compress the object detection model.Finally,perception results are fused using Hungarian al-gorithm,to achieve high-precision and low-cost vehicle road collaborative target perception in urban open road scenarios.Finally,experi-ments were conducted using a vehicle road collaborative dataset,and the results showed that the model achieved the detection ability of vehicle road collaborative perception in urban open road scenes,while verifying the feasibility and effectiveness of the algorithm proposed in this paper.