基于注意力机制的短道速滑运动轨迹预测模型

Short Track Speed Skating Trajectory Prediction Model Based on Attention Mechanism

为了解决短道速滑中多名运动员在拥挤状态下容易出现轨迹判断错误的问题,本文设计了一种基于注意力机制的轨迹预测模型。把位置和速度信息输入LSTM编码器中,再通过注意力模块对速度信息进行加权求和,最后整合速度和位置的隐藏状态输入到LSTM解码器来对短道速滑轨迹进行预测。结果采用平均位移误差(ADE)和最终位移误差(FDE)进行评估。结果表明:提出的轨迹预测模型在短道速滑运动员训练数据集中和在公开数据集中与基准模型相比平均ADE和FDE精度明显优于其他网络模型,具有一定的实用价值。

In order to solve the problem that many athletes are prone to wrong trajectory judgment in crowded state in short track speed skating,a trajectory prediction model based on attention mechanism is designed.This model inputs the position and speed information into the LSTM encoder,weights and sums the speed information through the attention module,and finally the hid⁃den state of the integrated speed and position is input to the LSTM Decoder to predict the short track speed skating trajectory.The results are evaluated by using average displacement error(ADE)and final displacement error(FDE).The results show that the av⁃erage ADE and FDE accuracy of the proposed track prediction model is significantly better than other network models in the train⁃ing dataset of short track speed skaters and in the open datasets,and has some practical value.