The invention relates to a U-turn intention and trajectory prediction method based on high-dimensional embedded depth parallel computing, which comprises the following steps of: shooting a video of a U-shaped U-turn position close to an intersection, extracting data, and generating a corresponding time sequence trajectory; performing U-turn intention analysis on the time sequence track of the vehicle on the left-turn lane, forming a new input by the straight-going vehicle on the left-turn lane, the U-turn vehicle and the straight-going vehicle on the opposite lane, and embedding behavior information of the vehicles into a high-dimensional space through an embedding layer; extracting time and space information of high-dimensional embedding, and combining the time and space information into spatio-temporal features; and inputting the spatial-temporal characteristic information into a lower-layer depth separable convolutional network, decoding the spatial-temporal characteristic information through a decoder, and outputting future trajectory information of all traffic participants. According to the technical scheme, the problems that an existing trajectory prediction method cannot accurately judge the turning-around intention of the vehicle, the vehicle trajectory prediction speed is low, and the accuracy is not high are effectively solved.
本发明涉及基于高维嵌入深度并行计算的掉头意图及轨迹预测方法,包括拍摄临近交叉口处U型掉头处的视频并提取数据,生成对应的时序轨迹;对左转车道上车辆的时序轨迹作掉头意图分析,并将左转车道上的直行车辆、掉头车辆以及对向车道的直行车辆组成新的输入,通过嵌入层将车辆的行为信息嵌入到高维空间中;提取高维嵌入的时间和空间信息,合并为时空特征;将时空特征信息输入到下层的深度可分离卷积网络中,通过解码器进行解码,输出所有交通参与者的未来轨迹信息。上述技术方案有效解决了现有轨迹预测方法无法对车辆掉头意图做出精确判断,以及车辆轨迹预测速度慢和准确率不高的问题。
Turn-around intention and trajectory prediction method based on high-dimensional embedded depth parallel computing
基于高维嵌入深度并行计算的掉头意图及轨迹预测方法
2023-06-23
Patent
Elektronische Ressource
Chinesisch
Driver intention prediction and intention trajectory updating method
Europäisches Patentamt | 2023
|Intention-Aware Dual Attention Based Network for Vehicle Trajectory Prediction
British Library Conference Proceedings | 2022
|