Driver Distraction Detection with a Two-stream Convolutional Neural Network 2020-01-1039
Driver distraction detection is crucial to driving safety when autonomous vehicles are co-piloted. Recognizing drivers’ behaviors that are highly related with distraction from real-time video stream is widely acknowledged as an effective approach mainly due to its non-intrusiveness. In recently years, deep learning neural networks have been adopted to by-pass the procedure of designing features artificially, which used to be the major downside of computer-vision based approaches. However, the detection accuracy and generalization ability is still not satisfying since most deep learning models extracts only spatial information contained in images. This research develops a driver distraction model based on a two-stream, spatial and temporal, convolutional neural network (CNN). The CNN in both stream is improved with Batch Normalization-Inception (BN-Inception) modules which increase the sparsity in the inception modules in GoogLeNet, so that the network is further speeded up and also more adapted to features at various-scales. The original RGB image is fed into the spatial stream CNN to extract static information, and the feature map of optical flow field extracted from adjacent image frames is fed into the temporal stream CNN to extract motion information. Based on a weighted fusion of two streams, the model outputs the classification result. Experimental results show that the proposed model recognize, with high accuracy, the driver’s hand and head movements and hence can predict the driver’s level of distraction with a good level of confidence.