| 摘 要: 为了快速检测车辆前排乘员是否系安全带，提出了一种改进YOLOv4-tiny网络及MobileNeXt网络的安全带检测算法。针对车辆及前排挡风玻璃的形状和大小，设计了一个结合深度可分离卷积、检测头可重构的轻量化YOLOv4-tiny目标检测框架，通过设置不同数量检测头的两个网络分别用于车辆和挡风玻璃检测。在MobileNeXt网络中，通过在沙漏残差模块中添加SAM空间注意力模块实现安全带检测。在车辆数据集、挡风玻璃数据集、安全带数据集上进行训练和测试，实验结果表明，车辆检测网络在平均精度(mean Average-Precision, mAP)为99.69%时速度为145 帧/秒，挡风玻璃检测网络在平均精度为99.88%时速度为163 帧/秒，安全带检测网络在准确率(Accuracy, ACC)为93.13%时速度为77 帧/秒。本文算法在兼顾速度的同时具有较高的检测精度。 |
| 关键词: 安全带检测；目标检测；深度可分离卷积 |
|
中图分类号: TP391.41
文献标识码: A
|
|
| Research on Seat Belt Detection Algorithm for Front Seat Occupants of Vehicles |
|
WANG Gangwei1, ZHANG Zhijia2
|
( 1.School of Information Science and Engineering, Shenyang University of Technology, Shenyang 110870, China;
2.School of Artificial Intelligence, Shenyang University of Technology, Shenyang 110870, China)
1611413651@qq.com; zhangzj@sut.edu.cn
|
| Abstract: In order to quickly detect whether the front seat occupants of the vehicle are wearing seat belts, this paper proposes a seat belt detection algorithm based on the improved YOLOv4-tiny network and MobileNeXt network. According to the shape and size of the vehicle and its front windshield, a lightweight YOLOv4-tiny target detection framework is designed that combines depthwise separable convolution and reconfigurable detection head. Two networks with different number of detection heads are set for detecting the vehicle and its windshield respectively. In the MobileNeXt network, seat belt detection is achieved by adding a SAM (Spatial Attention Module) to the hourglass residual module. Training and testing are performed on the vehicle dataset, windshield dataset, and seat belt dataset. Experimental results show that the vehicle detection network achieves a speed of 145 fps (frames per second) with a mean Average-Precision (mAP) of 99.69%. The windshield detection network achieves a speed of 163 fps with a mAP of 99.88%. The speed of the seat belt detection network is 77 fps with an accuracy (ACC) of 93.13%. The proposed algorithm has high detection accuracy while considering the speed. |
| Keywords: seat belt detection; object detection; depthwise separable convolution |
