Pose-aware Multi-position Feature Network for Driver Distraction Recognition

Abstract

This paper introduce pose-aware multi-position feature network for driver distraction recognition, taking into account the high association between the visual and geometric features of neighboring human key points. This method first uses the object detector to detect the driver’s body, and then uses the pose estimation to detect the key points of the human body. Finally, the key points of the human body are deconstructed to obtain key visual features and spatial features.In order to create the visual representation that corresponds to each key point, all the chosen visual representations are sent into a convolutional neural network. The geometric locations of key points on the upper body are used to build the spatial reasoning module, which is then transmitted to the neural network to get spatial features. The visual appearance features are fused with geometric features to predict corresponding driver action through linear layers.The experimental result shows that our methods achieve comparative performance on our own HY Large Vehicle Driver Dataset and the public AUC Driver Distracted Dataset.