Driver Detection Based on Deep Learning
Mingqi Lu, Yaocong Hu, Xiaobo Lu · Journal of Physics Conference Series · 2018
Driver detection is an essential part of monitoring and identifying drivers' bad driving behavior and preventing traffic accidents. The complicated background of the cockpit, the different light and dark light and the changeable driver pose bring great difficulties to the driver detection. In this paper, a driver detection algorithm based on improved Faster R-CNN is proposed based on RGB images captured by vehicle-mounted cameras. The residual structure is introduced into the ZF network to design the ZF-Net network, in order to improve the accuracy and maintain the real-time performance. LRN (Local Response Normalization) is replaced by BN (Batch Normalization), which simplifies parameter adjustment and accelerates network convergence. Experiments on a self-built driver image database demonstrate the effectiveness of the driver detection method.