Intelligent Vehicle Target Detection Algorithm Based on Multiscale Features.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Intelligent Transportation Systems

Background:

• Object detection models often face challenges with false and missed detections in complex intelligent driving scenarios.
• Existing models may have high computational loads and large sizes, limiting their real-time application.

Purpose of the Study:

• To optimize the YOLOv10 algorithm for enhanced object detection accuracy and reduced model complexity in intelligent driving.
• To develop a more efficient and effective detection framework for autonomous vehicles.

Main Methods:

• Designed Multi-Scale Flexible Convolution (MSFC) to capture simultaneous multi-scale information, reducing network depth and computational cost.
• Reconstructed the neck network using Shallow Auxiliary Fusion (SAF) and Advanced Auxiliary Fusion (AAF) for improved multi-scale feature extraction.
• Enhanced the detection head with multi-scale convolution and a channel adaptive attention mechanism for diverse and accurate feature extraction.

Main Results:

• The optimized YOLOv10 model achieved a file size of 13.4 MB, a reduction of 11.8% compared to the original.
• The model reached a mean Average Precision (mAP@0.5) of 93.0%, demonstrating superior detection accuracy.
• The improved model outperformed mainstream object detection models in overall performance, balancing accuracy and size.

Conclusions:

• The proposed enhancements significantly improve YOLOv10's performance for intelligent driving applications.
• This optimized model provides a practical solution for real-time object detection, balancing high accuracy with reduced computational resources.
• The developed framework offers a robust detection system for intelligent driving scenarios.