改进的轻量级GBC-YOLOv5疲劳驾驶检测方法

Improved Lightweight GBC-YOLOv5 Fatigue Driving Detection Method

针对现有的YOLO疲劳驾驶检测算法参数多、计算量大、实车推广困难等问题,提出一种改进的轻量级GBC-YOLOv5疲劳驾驶检测方法。首先,利用Ghost卷积模块实现特征提取;其次,采用Ghost Bottleneck模块来降低计算瓶颈,减少模型参数量和不必要的卷积计算;然后,引入CA注意力机制来增强网络的特征提取能力;最后,通过BiFPN结构来实现多尺度特征的高效融合。对YawDD疲劳驾驶数据集的检测结果表明,GBC-YOLOv5模型的平均精度均值达到了98.6%,实现了模型轻量化和实时性的平衡。

An improved lightweight GBC YOLOv5 fatigue driving detection method is proposed to address the problems of existing YOLO fatigue driving detection algorithms with multiple parameters,high computational complexity and difficulty in promoting actual vehicles.Firstly,feature extraction is achieved by using the Ghost convolution module.Secondly,the Ghost Bottleneck module is adopted to breakthrough computational bottlenecks,reducing model parameters and unnecessary convolution calculations.Then,CA attention mechanism is integrated to enhance the ability of network′s feature extraction.Finally,the BiFPN structure is introduced to achieve efficient multi-scale feature integration.The detection results of the YawDD fatigue driving dataset show that the average accuracy of the GBC YOLOv5 model reaches 98.6%,achieving a balance between model lightweight and real-time performance.