Synchronous End-to-End Vehicle Pedestrian Detection Algorithm Based on Improved YOLOv8 in Complex Scenarios.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems

Background:

• Traditional vehicle and pedestrian detection methods face limitations in complex urban traffic, including scale variation, occlusion, and high computational costs, hindering accuracy and real-time application.
• The demand for advanced dense pedestrian detection requires higher accuracy, reduced computational overhead, faster speeds, and easier deployment for applications like autonomous driving and public security.

Purpose of the Study:

• To develop an improved vehicle and pedestrian detection algorithm based on YOLOv8 to enhance detection accuracy and efficiency in complex traffic environments.
• To address the challenges of scale variation and severe occlusion in target detection.
• To achieve real-time detection capabilities suitable for intelligent transportation systems.

Main Methods:

• An improved YOLOv8 algorithm was developed, incorporating a deformable convolutional backbone network and an attention mechanism to optimize network structure.
• An end-to-end target search algorithm was introduced to enhance the stability and accuracy of vehicle and pedestrian detection.
• The improved model was evaluated on its precision, mean average precision (mAP), and frames per second (FPS) for real-time detection.

Main Results:

• The proposed algorithm achieved an 11.76% increase in precision and a 6.27% boost in mAP compared to baseline methods.
• The model demonstrated a real-time detection speed of 41.46 FPS, ensuring robust performance in complex scenarios.
• Applied to intelligent transportation systems, the improved YOLOv8 model reached a mAP of 95.23%, outperforming YOLOv5, YOLOv7, and Faster R-CNN.

Conclusions:

• The improved YOLOv8 algorithm significantly enhances the efficiency and robustness of vehicle and pedestrian detection, particularly in crowded urban settings.
• The developed model offers a superior solution for real-time detection in intelligent transportation systems, meeting the demands for higher accuracy and speed.
• The integration of deformable convolutions and attention mechanisms, along with an end-to-end target search, provides a robust framework for advanced traffic scene analysis.