Scholarlink

The Indian transportation landscape is unique, characterized by its non-laned traffic, a diverse mix of vehicles ranging from Motorized Two-wheelers(M2W), Motorized Three-wheelers (Auto-rickshaws), Cars, Light Commercial Vehicles (LCV), Buses, Trucks and Multi-Axle Trucks (MAT), and the spontaneous behavior of drivers. In response to India's rapid socio-economic advancement and urbanization, the nation has seen a significant increase in vehicle capacity. Urbanization rose from 23.10% in 1980 to 35.87% in 2022 [10], which reflects India's status as one of the world's fastest-growing countries. This urban population surge has led to more complex traffic behaviors, presenting both challenges and opportunities for traffic management. Existing systems fall short in handling this diversity and dynamism, leading to exacerbated traffic congestion, heightened accident rates, and inefficiencies in traffic management. Therein lies the necessity for an innovative solution that can not only comprehend but also predict and streamline the flow of vehicles. The rapid advancements in computer vision and artificial intelligence technologies have paved the way for novel approaches to tackle these issues. Introducing a transformative approach that leverages Unmanned Aerial Vehicles (UAVs) and cutting-edge deep learning models to pioneer a real-time vehicle detection and classification system tailored to the idiosyncrasies of Indian traffic conditions. Our study leverages You Only Look Once 8 (YOLOv8) for enhancing vehicle detection and classification within Indian traffic contexts, using a UAV-captured dataset with over 400 thousand annotations to train the model. This paper proposes an accurate, efficient, and real-time vehicle detection network based on the successful YOLOv8 object detection model. YOLOv8 model to the UAV dataset resulted in a mean average precision (mAP@0.5) of 0.756 and an F1 score of 0.86 across all vehicle classes. Compared to its predecessors, YOLOv8 demonstrated superior performance, advancing precision, recall, and overall accuracy for vehicle detection in India's complex traffic scenarios.

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

Development of Vehicle Detection and Classification System for Mixed Traffic Conditions Using Deep Learning

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Rajesh Chouhan,
  • Sanket Patil,
  • Ashish Dhamaniya

摘要

The Indian transportation landscape is unique, characterized by its non-laned traffic, a diverse mix of vehicles ranging from Motorized Two-wheelers(M2W), Motorized Three-wheelers (Auto-rickshaws), Cars, Light Commercial Vehicles (LCV), Buses, Trucks and Multi-Axle Trucks (MAT), and the spontaneous behavior of drivers. In response to India's rapid socio-economic advancement and urbanization, the nation has seen a significant increase in vehicle capacity. Urbanization rose from 23.10% in 1980 to 35.87% in 2022 [10], which reflects India's status as one of the world's fastest-growing countries. This urban population surge has led to more complex traffic behaviors, presenting both challenges and opportunities for traffic management. Existing systems fall short in handling this diversity and dynamism, leading to exacerbated traffic congestion, heightened accident rates, and inefficiencies in traffic management. Therein lies the necessity for an innovative solution that can not only comprehend but also predict and streamline the flow of vehicles. The rapid advancements in computer vision and artificial intelligence technologies have paved the way for novel approaches to tackle these issues. Introducing a transformative approach that leverages Unmanned Aerial Vehicles (UAVs) and cutting-edge deep learning models to pioneer a real-time vehicle detection and classification system tailored to the idiosyncrasies of Indian traffic conditions. Our study leverages You Only Look Once 8 (YOLOv8) for enhancing vehicle detection and classification within Indian traffic contexts, using a UAV-captured dataset with over 400 thousand annotations to train the model. This paper proposes an accurate, efficient, and real-time vehicle detection network based on the successful YOLOv8 object detection model. YOLOv8 model to the UAV dataset resulted in a mean average precision (mAP@0.5) of 0.756 and an F1 score of 0.86 across all vehicle classes. Compared to its predecessors, YOLOv8 demonstrated superior performance, advancing precision, recall, and overall accuracy for vehicle detection in India's complex traffic scenarios.