基于轻量化改进Yolov5s的驾驶员人脸检测算法

Driver face detection algorithm based on lightweight improved Yolov5s

人脸检测是驾驶员驾驶行为检测和驾驶疲劳检测的首要前提,而实时检测效率和检测精度则直接影响最终的判别效果,同时轻量化设计也更有助于后期在嵌入式设备上的计划部署;为此,基于Yolov5s设计了一种轻量化的驾驶员人脸检测算法。首先采用Shufflenetv2作为特征提取主干,降低网络的参数量和计算复杂度;然后引入ASFF实现不同层间的人脸信息融合,并结合SIoULoss提升模型训练速度和检测精度。结果表明,该检测算法的性能和轻量化程度均得到了有效的提升,节省了硬件成本,满足设计要求。

Face detection is a prerequisite for driver behavior detection and driving fatigue detection.Real-time detection efficiency and detection accuracy directly affect the final discrimination effect.At the same time,lightweight design is more conducive to the planned deployment on embedded devices in the later stage.Therefore,a lightweight driver face detection algorithm based on Yolov5s is designed.Firstly,Shufflenetv2 is used as the feature extraction backbone to reduce the network's parameter volume and computational complexity.Then,ASFF is introduced to achieve facial information fusion between different layers and combined with SIoU Loss to improve the model's training speed and detection accuracy.The results show that the performance and lightweight level of this detection algorithm have been effectively improved,which saves hardware costs and meets design requirements.