Scholarlink

<p>Accurate queue length estimation is critical for optimising signal control and evaluating performance at traffic intersections, especially in motorcycle-dominated mixed traffic in developing countries. This study proposes an integrated framework, Queue Length Estimation for Motorcycle-dominated Signalised Intersections (QLEMSI), that combines an AI-based object detection model, Harris corner detection, perspective transformation, and k-means clustering to address the unique challenges of vehicle occlusion and non-lane-based movements. A key innovation of the framework is the introduction of two “critical parameters” (Maximum Amplitude and Stability Duration) for identification of stationary vehicle platoons. The proposed method was tested on video data collected from a signalised intersection in Hanoi, Vietnam, validated using three datasets from different camera angles and junctions, and compared with other existing methods. The results demonstrated that QLEMSI achieved acceptable accuracy in estimating queue lengths, with a Root Mean Square Error (RMSE) as low as 1.09 zones (approximately 12&#xa0;m in an area of 60&#xa0;m before the stop line). The estimated number of queued zones closely matched manual ground-truth data, confirming the system’s potential for practical deployment in motorcycle-dominated mixed traffic signal optimisation. Given the extensive installation of surveillance cameras on signal pole arms, traffic agencies overseeing intersections dominated by motorcycles can effectively leverage the findings of this study. However, the first limitation is QLEMSI’s inability to identify the number of each vehicle type in queues, which is essential for traditional signal design. Another limitation is the need to validate zone lengths in different contexts. Additionally, night-time conditions are not considered.</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

Queue Length Estimation at Motorcycle-dominated Signalised Intersection by Using Computer Vision and K-Means Clustering

  • 期刊论文

  • 发表日期：

    2026年3月19日

查看全文
  • Tam Vu,
  • Ngoc Viet Pham,
  • Hong Nam Thai,
  • Ngoc Son Nguyen,
  • Tuan Anh Luong,
  • Van Minh Nguyen,
  • Viet Phuong Nguyen

摘要

Accurate queue length estimation is critical for optimising signal control and evaluating performance at traffic intersections, especially in motorcycle-dominated mixed traffic in developing countries. This study proposes an integrated framework, Queue Length Estimation for Motorcycle-dominated Signalised Intersections (QLEMSI), that combines an AI-based object detection model, Harris corner detection, perspective transformation, and k-means clustering to address the unique challenges of vehicle occlusion and non-lane-based movements. A key innovation of the framework is the introduction of two “critical parameters” (Maximum Amplitude and Stability Duration) for identification of stationary vehicle platoons. The proposed method was tested on video data collected from a signalised intersection in Hanoi, Vietnam, validated using three datasets from different camera angles and junctions, and compared with other existing methods. The results demonstrated that QLEMSI achieved acceptable accuracy in estimating queue lengths, with a Root Mean Square Error (RMSE) as low as 1.09 zones (approximately 12 m in an area of 60 m before the stop line). The estimated number of queued zones closely matched manual ground-truth data, confirming the system’s potential for practical deployment in motorcycle-dominated mixed traffic signal optimisation. Given the extensive installation of surveillance cameras on signal pole arms, traffic agencies overseeing intersections dominated by motorcycles can effectively leverage the findings of this study. However, the first limitation is QLEMSI’s inability to identify the number of each vehicle type in queues, which is essential for traditional signal design. Another limitation is the need to validate zone lengths in different contexts. Additionally, night-time conditions are not considered.