{"id":194,"date":"2025-01-23T10:19:53","date_gmt":"2025-01-23T15:19:53","guid":{"rendered":"https:\/\/ise.ncsu.edu\/icel\/?page_id=194"},"modified":"2025-06-02T15:07:50","modified_gmt":"2025-06-02T19:07:50","slug":"pedestrian-behavior-prediction-models","status":"publish","type":"page","link":"https:\/\/ise.ncsu.edu\/icel\/research\/pedestrian-behavior-prediction-models\/","title":{"rendered":"Pedestrian Behavior Prediction Models"},"content":{"rendered":"\n
\n
<\/div>\n\n\n\n
\n

Pedestrian Behavior Prediction Models<\/h2>\n<\/div>\n\n\n\n
<\/div>\n<\/div>\n\n\n\n

Last Updated<\/strong>: 01\/23\/2025 | All information is accurate and up-to-date<\/p>\n\n\n\n

\n
\n
<\/div>\n\n\n\n
\n

Background and Objectives<\/strong><\/h3>\n\n\n\n
    \n
  • In most cases, automated vehicles (AVs) can drive smoothly on highways and freeways.<\/li>\n\n\n\n
  • However, AVs still face challenges when it comes to driving in urban settings\n
      \n
    • One key challenge: \u201cUnpredictable\u201d and rapidly changing behaviors of pedestrians and vulnerable road users<\/li>\n\n\n\n
    • The existing driving strategy is over-conservative targeting to avoid crashes based on short-term kinematics calculations<\/li>\n<\/ul>\n<\/li>\n\n\n\n
    • The main research objective is to better predict pedestrian behaviors with deep-learning algorithms to support driving decision-making during pedestrian encounters in complex road scenes<\/li>\n<\/ul>\n<\/div>\n\n\n\n
      <\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n
      \n
      <\/div>\n\n\n\n
      \n

      Two-Tower Ego-Centric Pedestrian Trajectory Prediction<\/h3>\n\n\n\n

      Key Features<\/strong><\/p>\n\n\n\n

        \n
      1. Multi-modal inputs;<\/li>\n\n\n\n
      2. Two-tower model to decompose egocentric pedestrian trajectories based on ego-vehicle and pedestrian movements;<\/li>\n\n\n\n
      3. Inferences of pedestrian future moving directions.<\/li>\n<\/ol>\n\n\n
        \n
        \"A
        Model Structure<\/figcaption><\/figure>\n<\/div>\n\n
        \n
        \"Multiple
        Prediction Results at 1.5 Seconds<\/figcaption><\/figure>\n<\/div>\n\n\n

        Algorithm Results on JAAD Benchmark Dataset<\/strong><\/p>\n\n\n\n

        Method<\/strong><\/td>Average Displacement Error<\/strong><\/td>Final Displacement Error<\/strong><\/td><\/tr>
        PIE<\/td>22.83<\/td>49.44<\/td><\/tr>
        BiPed<\/td>21.13<\/td>48.88<\/td><\/tr>
        Two-Tower Model<\/td>17.92<\/strong><\/td>41.33<\/strong><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

        PIE: <\/strong>Rasouli, A., Kotseruba, I., Kunic, T. and Tsotsos, J.K., 2019. Pie: A large-scale dataset and models for pedestrian intention estimation and trajectory prediction. In Proceedings of the IEEE\/CVF International Conference on Computer Vision (pp. 6262-6271).<\/p>\n\n\n\n

        BiPed<\/strong>: Rasouli, A.; Rohani, M.; and Luo, J. 2021. Bifold and Semantic Reasoning for Pedestrian Behavior Prediction. In Proceedings of the IEEE\/CVF International Conference on Computer Vision (ICCV), 15600\u201315610.<\/p>\n\n\n\n

        Algorithm Results on PSI Benchmark Dataset<\/strong><\/p>\n\n\n\n

        Method<\/strong><\/td>Average Displacement Error<\/strong><\/td>Final Displacement Error<\/strong><\/td><\/tr>
        PIE<\/td>35.39<\/td>61.50<\/td><\/tr>
        Two-Tower Model<\/td>22.34<\/strong><\/td>46.63<\/strong><\/td><\/tr><\/tbody><\/table><\/figure>\n<\/div>\n\n\n\n
        <\/div>\n<\/div>\n\n\n\n
        \n
        \n
        <\/div>\n\n\n\n
        \n

        Dual-View Pedestrian Trajectory Prediction<\/h3>\n\n\n\n

        Key Features<\/strong><\/p>\n\n\n\n

          \n
        1. Multi-modal inputs;<\/li>\n\n\n\n
        2. Predicting bird\u2019s eye trajectory, ego-centric trajectory, and pedestrian actions simultaneously.<\/li>\n\n\n\n
        3. Multi-task learning to improve prediction accuracy.<\/li>\n<\/ol>\n\n\n
          \n
          \"A
          Model Structure<\/figcaption><\/figure>\n<\/div>\n\n
          \n
          \"Multiple
          Prediction Results at 2 Seconds<\/figcaption><\/figure>\n<\/div>\n\n\n

          Algorithm Results on nuScenes Benchmark Dataset<\/strong><\/p>\n\n\n\n

          Method<\/strong><\/td>Bitrap<\/strong><\/td>SGNet<\/strong><\/td>Ours<\/strong><\/td><\/tr>
          Average Displacement Error<\/strong>
          Bird\u2019s eye view<\/td>          		49<\/td>46<\/td>28<\/strong><\/td><\/tr>
          Final Displacement Error<\/strong>
          Bird\u2019s eye view<\/td>          		57<\/td>55<\/td>41<\/strong><\/td><\/tr>
          Average Displacement Error<\/strong>
          Egocentric view<\/td>          		92<\/td>89<\/td>61<\/strong><\/td><\/tr>
          Final Displacement Error<\/strong>
          Egocentric view<\/td>          		112<\/td>102<\/td>86<\/strong><\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n

          Bitrap<\/strong>: <\/strong>Yao, Y., Atkins, E., Johnson-Roberson, M., Vasudevan, R. and Du, X., 2021. Bitrap: Bi-directional pedestrian trajectory prediction with multi-modal goal estimation. IEEE Robotics and Automation Letters, 6(2), pp.1463-1470.<\/p>\n\n\n\n

          SGNet<\/strong>: Wang, C., Wang, Y., Xu, M. and Crandall, D.J., 2022. Stepwise goal-driven networks for trajectory prediction. IEEE Robotics and Automation Letters<\/em>, 7<\/em>(2), pp.2716-2723.<\/p>\n<\/div>\n\n\n\n

          <\/div>\n<\/div>\n<\/div><\/div>\n\n\n\n
          \n
          <\/div>\n\n\n\n
          \n

          Pedestrian Intention Prediction Models<\/h3>\n\n\n\n

          Key Features of VR-GCN<\/strong><\/p>\n\n\n\n

            \n
          1. Scene graph with 32 objects;<\/li>\n\n\n\n
          2. CNN + GCN + LSTM as the main structure;<\/li>\n\n\n\n
          3. Pose information incorporated;<\/li>\n<\/ol>\n\n\n
            \n
            \"\"
            Graph-Convolutional-Network-based Pedestrian Intent Prediction (VR-GCN)<\/figcaption><\/figure>\n<\/div>\n\n\n

            Key Features of TrEP<\/strong><\/p>\n\n\n\n

              \n
            1. Transformer-based feature extraction and encoding;<\/li>\n\n\n\n
            2. Evidential learning for robust performance and calibrated uncertainty.<\/li>\n<\/ol>\n\n\n
              \n
              \"\"
              Transformer-based Evidential Learning for Intent Prediction (TrEP)<\/figcaption><\/figure>\n<\/div>\n\n\n

              Algorithm Results on nuScenesBenchmark Dataset<\/strong><\/p>\n<\/div>\n\n\n\n

              <\/div>\n<\/div>\n\n\n\n
              Method<\/strong><\/td>Accuracy<\/strong><\/td>Balanced Accuracy<\/strong><\/td>F1<\/strong><\/td><\/tr>
              VR-GCN<\/strong><\/td>0.74<\/td>0.61<\/td>0.64<\/td><\/tr>
              TrEP<\/strong><\/td>0.85<\/td>0.77<\/td>0.90<\/td><\/tr><\/tbody><\/table><\/figure>\n\n\n\n
              \n
              \n
              <\/div>\n\n\n\n
              \n

              Intention Prediction Demo<\/h3>\n\n\n\n
              \n