Scholarlink

<p>For a traditional traffic situational awareness system (TSAS), its “Road-side unit (RSU) + cloud-based analysis” structure is difficult to meet the demands of rapidly expanding urban areas. Relatively high costs of microwave speed detection modules and bandwidth requirements of information systems significantly increase construction costs. By computer vision (CV) and edge computing technologies, traffic situational awareness tasks can be integrated into cheaper edge devices (roadside surveillance, RSS), effectively addressing such challenges. In this study, we present a low-cost TSAS developed based on YOLO v8 and grey wolf optimizer-long short-term memory (GWO-LSTM) neural network. Proposed system can automatically perform vehicle and license plate recognition, speed measurement, and data recording within the field of view of RSSs. Additionally, it accurately predicts the future traffic conditions of monitored roads using recorded information. Experimental results demonstrate that the proposed TSAS achieves a license plate recognition accuracy of 97.7%, vehicle type recognition accuracy of 98.1%, and speed measurement error of less than 0.45&#xa0;km/h, with R2 of 0.8971 for GWO-LSTM predictions. This system is sufficiently effective for traffic monitoring and situational awareness tasks but enforcement forensic applications.</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

Low-cost real-time traffic situational awareness system based on modified YOLO v8 and GWO-LSTM for edge deployment

  • 期刊论文

  • 发表日期：

    2025年3月25日

查看全文
  • Jianwen Liu,
  • Ruyue Gong,
  • Yi Gong,
  • Zeqin Li,
  • Zhiwei Chen

摘要

For a traditional traffic situational awareness system (TSAS), its “Road-side unit (RSU) + cloud-based analysis” structure is difficult to meet the demands of rapidly expanding urban areas. Relatively high costs of microwave speed detection modules and bandwidth requirements of information systems significantly increase construction costs. By computer vision (CV) and edge computing technologies, traffic situational awareness tasks can be integrated into cheaper edge devices (roadside surveillance, RSS), effectively addressing such challenges. In this study, we present a low-cost TSAS developed based on YOLO v8 and grey wolf optimizer-long short-term memory (GWO-LSTM) neural network. Proposed system can automatically perform vehicle and license plate recognition, speed measurement, and data recording within the field of view of RSSs. Additionally, it accurately predicts the future traffic conditions of monitored roads using recorded information. Experimental results demonstrate that the proposed TSAS achieves a license plate recognition accuracy of 97.7%, vehicle type recognition accuracy of 98.1%, and speed measurement error of less than 0.45 km/h, with R2 of 0.8971 for GWO-LSTM predictions. This system is sufficiently effective for traffic monitoring and situational awareness tasks but enforcement forensic applications.