融合坐标注意力与卷积社交池的车辆轨迹预测

Vehicle trajectory prediction integrating coordinate attention and convolutional social pool

针对车辆行驶受相邻车辆的影响,本文提出了一种特征增强的模型,在LSTM编码器-解码器架构上,使用卷积社交池提取一定范围内的车辆轨迹信息特征,再添加坐标注意力进行特征增强。使用NGSIM数据集实验结果表明,本文所提模型表现性能良好。相较于其他模型,5 s内在均方根误差指标上平均降低了14.10%、15.76%、17.49%、17.81%、17.20%。

Addressing the impact of neighboring vehicles on vehicle motion,this paper introduces a feature-enhanced model.Within the LSTM encoder-decoder architecture,it employs convolutional social pooling to extract vehicle trajectory information features within a specific range and further enhances these features with coordinate attention.Experimental results using the NGSIM dataset demonstrate the excellent performance of the proposed model.Compared to other models,it achieves an average reduction of 14.10%,15.76%,17.49%,17.81%,and 17.20%in the root mean square error(RMSE)metric within a 5-second prediction horizon.