基于注意力机制的车辆行为预测

Vehicle behavior prediction based on attention mechanism

为了解决智能驾驶场景中对周边车辆未来行为的预测问题,研究了基于注意力机制的长短时记忆网络(LSTM)模型的车辆行为预测方法.首先提出了一种非均匀步长的时间序列数据划分方法,将属于特定行为的车辆时序信息进行分类;以LSTM为基本的神经网络框架,用注意力机制判断输入时序信息中各个时间步信息的重要程度,分配不同的权重值;以目标车辆及其周边车辆的历史轨迹信息作为算法输入,用来预测目标车辆将来的运动行为.结果表明:该算法可以解决固定步长的时序分类方法导致的信息遗漏或计算资源负担增加的问题,同时能够有效提高行为预测准确性,减少车辆行为预测时间.

To predict the future behavior of surrounding vehicles in intelligent driving scenarios,the vehicle behavior prediction method was investigated based on the attention mechanism of long and short-term memory(LSTM)network model.The time series data division method of non-uniform step-size was proposed to classify the vehicle time series information of specific behaviors.With LSTM as the basic neural network framework,the attention mechanism was used to assign different weight values for each time step according to the importance in the time series information.The historical trajectory information of the target vehicle and its surrounding vehicles was used as algorithm input to predict the future motion behavior of the target vehicle.The results show that the algorithm can solve the problems of information omission and increasing of computing resource burden caused by the fixed step-size classification method.The algorithm can effectively improve the accuracy of behavior prediction and reduce the time using of vehicle behavior prediction.