基于通道注意力YOLOV5s的驾驶行为识别研究

设计了一种集成通道注意力机制的YOLOV5s检测网络的驾驶行为识别方法,用以实时检测并识别驾驶员在驾驶室内的驾驶行为,从而有利于纠正驾驶员的不良驾驶行为,减少交通事故发生的概率。建立了驾驶室内驾驶员手部动作的图像数据集;在YOLOv5s网络结构中引入通道注意力机制,通过对比实验、消融实验研究了通道注意力模块嵌入YOLOv5s中的较佳作用位置、配置数量的影响及其检测识别性能效果;论证了带通道注意力的改进YOLOV5s可保留信息量大的特征、抑制不相关的特征,模型参数量和复杂度降低,从而加快检测速度。测试结果显示,较原YOLOV5s网络,改进的YOLOV5s在平均精确度和召回率上相当,而检测速度提升了26.08%,该方法能够较好地满足驾驶员手部动作的实时监控需求。

Study on Identification of driving behavior based on YOLOV5s with channel attention

A driving behavior identification method of YOLOV5s detection network with integrated channel attention mechanism is designed to detect and identify the driver"s driving behavior in the cab in real time, so as to correct the driver"s bad driving behavior and reduce the probability of traffic accidents. The image data set of the hand movements of the driver in the cab is established, the channel attention mechanism is introduced into the YOLOv5s network structure, the effect of configuration quantity and the improved YOLOV5s with channel attention can retain the features of large information and suppress irrelevant features, and reduce the number and complexity of model parameters to accelerate the detection speed. The test results show that compared with the original YOLOV5s network, the improved YOLOV5s is comparable in average accuracy and recall rate, while the detection speed is increased by 26.08%. This method can better meet the real-time monitoring requirements of drivers" hand movements.