车辆检测中YOLO模型的综合性能评估与实证分析

Comprehensive Performance Evaluation and Empirical Analysis of YOLO Models for Vehicle Detection Evaluation

摘要: 针对交通路口复杂场景中车辆检测存在的实时性与精度难以兼顾、适应性不足的问题，本文提出一种融合精度、召回率及检测帧率的加权综合评价指标。基于VisDrone数据集，系统评估YOLOv5至YOLOv10等主流模型，并分析图像清晰度提升与掩模策略的优化效果。实验结果表明，YOLOv10n在综合指标得分最高，达150.41；提高图像像素使综合性能提升幅度达23.6%，显著优于单纯增加模型复杂度。研究为智能交通系统中的车辆检测提供了高效且适应性强的技术支持。

Abstract: To address the challenges of balancing real-time performance and detection accuracy, as well as insufficient adaptability in complex traffic intersection scenarios, this study proposes a weighted comprehensive evaluation metric that integrates precision, recall, and detection frame rate. Based on the VisDrone dataset, mainstream models from YOLOv5 to YOLOv10 are systematically evaluated, and optimization strategies such as image clarity enhancement and mask application are analyzed. Experimental results show that YOLOv10n achieves the highest comprehensive score of 150.41. Enhancing image resolution improves overall performance by 23.6%, which is significantly more effective than simply increasing model complexity. This research provides an efficient and adaptable technical solution for vehicle detection in intelligent transportation systems.