Scholarlink

<p>In today’s rapidly evolving world, ensuring safe navigation in environments where pedestrians and vehicles exist together is a challenge for self-driving vehicles. Pedestrians exhibit highly unpredictable movement patterns influenced by crossings, group interactions, sudden stops, speed changes, and various environmental factors, making it extremely challenging for self-driving vehicles to accurately predict paths and prevent collisions. Traditional motion prediction models based on rules of physics have somehow been able to capture the motion of humans, but have not gained good accuracy. While some previous models have gained better accuracy in this topic using deep learning and other machine learning models, there is still room for improvement. A reliable and efficient pedestrian motion prediction system is crucial for autonomous vehicle safety and seamless human-vehicle interaction in smart cities. To deal with this issue, we propose Pedestrian Motion Prediction for Autonomous Vehicle Navigation (PMPAVN), a machine learning-based approach designed to predict pedestrian trajectories with high accuracy and precision. This model enables autonomous vehicles to predict pedestrian movement in real time, allowing for instant decision-making and collision avoidance and enhances prediction accuracy. The model integrates a novel approach that combines Bi-LSTM with SAM2 for attention, enabling enhanced adaptability in dynamic environments. Experimental results show that PMPAVN outperforms existing pedestrian motion prediction models, achieving an ADE of 0.58&#xa0;m and FDE of 1.27&#xa0;m, improving over other models like S-LSTM and Arc-LSTM-SMF. Additionally, PMPAVN significantly enhances computational efficiency compared to conventional LSTM models, making it well-suited for real-time autonomous navigation.</p>

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

Pedestrian Motion Prediction for Autonomous Vehicle Navigation Using Machine Learning

  • 期刊论文

  • 发表日期：

    2026年1月30日

查看全文
  • Sanjit Kumar Dash,
  • Jyotirmayee Das,
  • Suvesh Samantaray,
  • Muktikanta Sahu

摘要

In today’s rapidly evolving world, ensuring safe navigation in environments where pedestrians and vehicles exist together is a challenge for self-driving vehicles. Pedestrians exhibit highly unpredictable movement patterns influenced by crossings, group interactions, sudden stops, speed changes, and various environmental factors, making it extremely challenging for self-driving vehicles to accurately predict paths and prevent collisions. Traditional motion prediction models based on rules of physics have somehow been able to capture the motion of humans, but have not gained good accuracy. While some previous models have gained better accuracy in this topic using deep learning and other machine learning models, there is still room for improvement. A reliable and efficient pedestrian motion prediction system is crucial for autonomous vehicle safety and seamless human-vehicle interaction in smart cities. To deal with this issue, we propose Pedestrian Motion Prediction for Autonomous Vehicle Navigation (PMPAVN), a machine learning-based approach designed to predict pedestrian trajectories with high accuracy and precision. This model enables autonomous vehicles to predict pedestrian movement in real time, allowing for instant decision-making and collision avoidance and enhances prediction accuracy. The model integrates a novel approach that combines Bi-LSTM with SAM2 for attention, enabling enhanced adaptability in dynamic environments. Experimental results show that PMPAVN outperforms existing pedestrian motion prediction models, achieving an ADE of 0.58 m and FDE of 1.27 m, improving over other models like S-LSTM and Arc-LSTM-SMF. Additionally, PMPAVN significantly enhances computational efficiency compared to conventional LSTM models, making it well-suited for real-time autonomous navigation.