基于YOLOv7-DCA的疲劳检测方法研究

Research on Fatigue Detection Method Based on YOLOv7-DCA

针对疲劳检测中小尺度检测效果不佳和实时性差等问题,以矿井提升机司机疲劳检测为目标,对YOLOv7的结构进行精简并且基于AIoU(Area Intersection over Union)损失函数优化预测框与验证框的回归过程.在模型中引入双通道注意力机制实现小尺度特征的信息增强,通过融合眨眼频率、闭眼时长和打哈欠时长来判断司机的状态.实验结果表明,本文方法对疲劳检测精度达到98.85%,检测速度达到70 FPS,与其他算法相比,本文算法具有更好的准确性和实时性.

Addressing issues of poor performance and low real-time effectiveness in small-scale fatigue detection,with a focus on fatigue detection for mine hoist operators,this paper streamlines the structure of YOLOv7 and optimizes the regression process of prediction boxes and validation boxes based on the AIoU(Area Intersection over Union)loss function,introducing a dual-channel attention mechanism in the model enhances information for small-scale features,determining the driver's state by combining blink frequency,closure duration and yawn duration.Experimental results indicate that the accuracy of fatigue detection in this paper reaches 98.85%,and the speed reaches 70FPS.Compared to other algorithms,the algorithm proposed in this paper demonstrates better accuracy and real-time performance.