Scholarlink

To enhance traffic safety in urban road merging areas, this study introduces a traffic conflict prediction method utilizing a Multi-Head CNN-LSTM Attention (CLMHA) trajectory prediction model. The multi-head CNN-LSTM attention model integrates the characteristics of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, employing a multi-head attention mechanism to effectively extract vehicle trajectory features. This approach is particularly adept at addressing the spatial dependencies and temporal dynamics inherent in vehicle trajectory prediction within merging areas characterized by complex environments. To mitigate redundancy in feature data and to obtain complementary features, a convolutional neural network feature fusion technique is proposed, facilitating improved extraction of the spatial and temporal motion relationships of vehicles. The efficacy of the method is validated using the Mirror-Traffic dataset. The results indicate that the proposed trajectory prediction algorithm demonstrates superior accuracy in forecasting future vehicle trajectories compared to alternative algorithms, thereby yielding enhanced outcomes in traffic conflict prediction.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Traffic Conflict Research Approach Based on Multi-head CNN-LSTM Attention Trajectory Prediction

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Meng Ren,
  • Haibo Zhang,
  • Jun Fang

摘要

To enhance traffic safety in urban road merging areas, this study introduces a traffic conflict prediction method utilizing a Multi-Head CNN-LSTM Attention (CLMHA) trajectory prediction model. The multi-head CNN-LSTM attention model integrates the characteristics of Convolutional Neural Networks (CNN) and Long Short-Term Memory (LSTM) networks, employing a multi-head attention mechanism to effectively extract vehicle trajectory features. This approach is particularly adept at addressing the spatial dependencies and temporal dynamics inherent in vehicle trajectory prediction within merging areas characterized by complex environments. To mitigate redundancy in feature data and to obtain complementary features, a convolutional neural network feature fusion technique is proposed, facilitating improved extraction of the spatial and temporal motion relationships of vehicles. The efficacy of the method is validated using the Mirror-Traffic dataset. The results indicate that the proposed trajectory prediction algorithm demonstrates superior accuracy in forecasting future vehicle trajectories compared to alternative algorithms, thereby yielding enhanced outcomes in traffic conflict prediction.