X.J. Guo, H. Cai, M. Li

Pages: 171-180

Abstract
The recognition of driver's behavior state is of great significance to the safety of motor vehicle driving. In order to improve the recognition rate and reduce the omission rate, a deep learning based recognition method was proposed for drivers unwearing seat belts during vehicle driving. Firstly, after the segmentation of the driver's skin color, the driver image is enhanced by maximizing local changes. Secondly, combined with the results of license plate region location and window region location, the driver region is recognised. Finally, the driver's area is taken as the input variable of the model, and the result of the recognition of the unwearing seat belt is taken as the output variable. The recognition model of the unwearing seat belt is constructed using the convolution neural network in the deep learning. The experimental results show that the maximum recognition rate of this method is 96%, and the maximum miss detection rate is only 0.58%.
Keywords: deep learning; vehicle driving process; driver; unwearing safety belt; convolution neural network; loss function