Scholarlink

In this paper, we present a novel digital twin framework for pedestrian analysis in autonomous driving. The digital twin replicates real-world systems, enabling the monitoring, simulation, and analysis of pedestrian detection algorithms in a controlled environment. Our framework integrates the YOLOv9 model on a Raspberry Pi, enhanced by a Movidius Neural Compute Stick, to detect pedestrians and transmit data via MQTT to the CARLA Simulator. Additionally, we have integrated MediaPipe with YOLOv9 to detect pedestrian poses, such as walking, standing, or unknown, and mirrored these behaviors in CARLA. If discrepancies are detected, feedback is sent to the Raspberry Pi to enable potential corrections. This setup allows extensive testing across diverse scenarios, improving the safety and reliability of autonomous vehicles. To validate the proposed approach, we developed a use case and compared execution times for the detection algorithm on Raspberry Pi 5, with and without the Movidius Neural Compute Stick. Our study details the system architecture and implementation, highlighting significant advancements in pedestrian analysis for autonomous driving technologies.

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

Autonomous Driving Pedestrian Analysis: A Digital Twin Approach Using Raspberry Pi and CARLA Simulator via MQTT

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Narmada Ambigapathy,
  • Fatima Idrees,
  • Katrin Glöwing,
  • Charles Steinmetz,
  • Achim Rettberg

摘要

In this paper, we present a novel digital twin framework for pedestrian analysis in autonomous driving. The digital twin replicates real-world systems, enabling the monitoring, simulation, and analysis of pedestrian detection algorithms in a controlled environment. Our framework integrates the YOLOv9 model on a Raspberry Pi, enhanced by a Movidius Neural Compute Stick, to detect pedestrians and transmit data via MQTT to the CARLA Simulator. Additionally, we have integrated MediaPipe with YOLOv9 to detect pedestrian poses, such as walking, standing, or unknown, and mirrored these behaviors in CARLA. If discrepancies are detected, feedback is sent to the Raspberry Pi to enable potential corrections. This setup allows extensive testing across diverse scenarios, improving the safety and reliability of autonomous vehicles. To validate the proposed approach, we developed a use case and compared execution times for the detection algorithm on Raspberry Pi 5, with and without the Movidius Neural Compute Stick. Our study details the system architecture and implementation, highlighting significant advancements in pedestrian analysis for autonomous driving technologies.