[论文解读] Real-time Multi-Class Helmet Violation Detection Using Few-Shot Data Sampling Technique and YOLOv8

Armstrong Aboah, Bin Wang+2 more|arXiv (Cornell University)|Apr 13, 2023

Traffic and Road Safety被引用 15

一句话总结

本文提出一个使用新颖的 few-shot 数据采样策略与 YOLOv8 的实时头盔违规检测系统，在 AI City Challenge Track 5 中实现具有竞争力的 mAP 和实时推理。

ABSTRACT

Traffic safety is a major global concern. Helmet usage is a key factor in preventing head injuries and fatalities caused by motorcycle accidents. However, helmet usage violations continue to be a significant problem. To identify such violations, automatic helmet detection systems have been proposed and implemented using computer vision techniques. Real-time implementation of such systems is crucial for traffic surveillance and enforcement, however, most of these systems are not real-time. This study proposes a robust real-time helmet violation detection system. The proposed system utilizes a unique data processing strategy, referred to as few-shot data sampling, to develop a robust model with fewer annotations, and a single-stage object detection model, YOLOv8 (You Only Look Once Version 8), for detecting helmet violations in real-time from video frames. Our proposed method won 7th place in the 2023 AI City Challenge, Track 5, with an mAP score of 0.5861 on experimental validation data. The experimental results demonstrate the effectiveness, efficiency, and robustness of the proposed system.

研究动机与目标

• 推动在多样天气和光照条件下研发一个鲁棒、实时的头盔违规检测系统。
• 通过引入一个 few-shot 数据采样框架来减轻标注负担。
• 评估 YOLO 家族模型以确定头盔违规的最佳实时检测器。

提出的方法

• 引入一个 few-shot 数据采样技术以选择具有代表性的帧并减少标注工作量。
• 应用数据增强以提升训练多样性以及对遮挡和视角变化的鲁棒性。
• 使用测试时增强（TTA）以提升推理精度。
• 比较 YOLOv5、YOLOv7 和 YOLOv8 以识别该任务中表现最佳的单阶段检测器。
• 在 RTX 3090 上以 400 轮训练、批量大小 16、图像大小 832x832 训练模型。
• 纳入背景帧和负样本以降低误报。

Figure 1 : Illustration of predictions of motorists, passengers, and motorcycles in real-time. We present a method that is able to predict helmet rule violation in real-time. The different bounding box colors represent different predictive class.

实验结果

研究问题

• RQ1 few-shot 数据采样方法在降低标注需求的同时保持检测性能的效果如何？
• RQ2 哪一种 YOLO 单阶段检测器（v5/v7/v8）在实时头盔违规检测中提供最佳的精度与速度平衡？
• RQ3 测试时增强是否在多样天气和光照条件下提升检测性能？

主要发现

模型	mAP.05	mAP.05-.95	精度	召回率
yolov5	0.823	0.465	0.892	0.811
yolov7	0.846	0.526	0.912	0.854
yolov8	0.858	0.601	0.923	0.898
yolov5+TTA	0.911	0.613	0.931	0.907
yolov8+TTA	0.923	0.647	0.953	0.918

• YOLOv8+TTA 模型在验证集上实现了最高的 validation mAP.05-.95，为 0.647。

Figure 2 : Images from different time of day and weather. a) Daylight, b) Nighttime, and c)Foggy weather.