The invention discloses a deep learning model for modeling left-turn vehicle behaviors in a mixed flow intersection scene where potential conflicts exist between vehicles and electric bicycles. The method mainly comprises the following steps: preprocessing data, inputting a historical track of a vehicle into a vehicle motion encoder, and extracting historical track features of the vehicle; inputting a relative position from the vehicle to the potential target point into a driving intention extraction module, introducing a maximum pooling layer, and capturing a driving intention feature going to the potential target point; the outputs of the first two modules are input into a target flow interaction module for modeling interaction between target flow vehicles; inputting the historical trajectory of the mixed flow object and the relative position from the vehicle to the mixed flow object into a mixed flow interaction module; and the extracted features are synthesized, and a future driving track of the vehicle is generated through decoding.
本发明公开了一种在车辆与电动自行车存在潜在冲突的混合流交叉口场景下建模左转车辆行为的深度学习模型。主要包括以下步骤:对数据进行预处理,将车辆历史轨迹输入车辆运动编码器,提取车辆的历史轨迹特征;将从车辆到潜在目标点的相对位置输入驾驶意图提取模块,引入最大池化层,捕获前往潜在目标点的驾驶意图特征;再将前两个模块的输出输入目标流交互模块,用于建模目标流车辆之间的交互;将混合流对象历史轨迹和从车辆到混合流对象的相对位置输入混合流交互模块;再综合上述提取所得的特征,解码生成车辆未来的行驶轨迹。
Deep learning modeling method for interaction behaviors of left-turn vehicles and illegal bicycles at intersection
交叉口左转车与违规自行车交互行为的深度学习建模方法
2024-06-11
Patent
Electronic Resource
Chinese
The Impact Analysis of Signal Intersection Right-Turn Motor Vehicles on Straight Bicycles
British Library Conference Proceedings | 2013
|The Impact Analysis of Signal Intersection Right-Turn Motor Vehicles on Straight Bicycles
Trans Tech Publications | 2012
|Illegal parking identification method for shared bicycles
European Patent Office | 2023
|