摘要
对周围环境中运动物体未来状态的准确预测是影响自动驾驶车辆做出准确决策的重要影响因素,车辆是最常见也是最需要关注的运动物体之一;针对结构化道路下周围车辆轨迹预测的多模态输入问题,提出了基于注意力机制的深度预测网络;提出交互模块以提取目标车辆与周围车辆及车道线信息存在的交互特征;结合车道线信息对车辆运动的指引作用,加入目标点预测模块以预测目标车辆可能到达的目标点,增加预测准确性;在Argoverse公开数据集上进行实验,所提轨迹预测网络在3秒预测时长实现了1.45 m最小平均距离误差及3.21 m最小最终距离误差的预测精度,优于当前主流的预测算法。
The accurate prediction of the future state of moving objects in the surrounding environment is an important influencing factor for autonomous vehicles to make accurate decisions.A vehicle is one of the most common and demanding moving objects.A depth prediction network based on attention mechanism is proposed for the multi-modal input problem of vehicle trajectory prediction on the structural road.An interactive module is proposed to extract the interactive features of target vehicle,surrounding vehicles and lane line information.Combined with the guidance of lane line information to the vehicle movement,the target point prediction module is added to predict the possible target point of target vehicle to increase the prediction accuracy.The experiments are conducted on Argoverse public dataset,the proposed prediction network achieves the prediction accuracy for the minimum average distance error of 1.45 m and minimum final distance error of 3.21 m in the prediction range of 3 s,which is better than current mainstream methods.
作者
刘建铭
陈伟侠
卢仲康
LIU Jianming;CHEN Weixia;LU Zhongkang(School of Mechanical and Automotive Engineering,South China University of Technology,Guangzhou 510641,China;Guangzhou Huagong Automobile Inspection Technology Co.,Ltd.,Guangzhou510640,China)
出处
《计算机测量与控制》
2023年第11期106-112,118,共8页
Computer Measurement &Control
关键词
自动驾驶
轨迹预测
深度学习
门控循环单元
注意力机制
autonomous driving
trajectory prediction
deep learning
gated recurrent unit(GRU)
attention mechanism
