Pose-guided model for driving behavior recognition using keypoint action learning

The number of traffic accidents caused by distracted driving is increasing worldwide. There is an urgent need for a system for detection and warning of distracted driving. Therefore, we propose a pose-guided model using the keypoint action features to recognize driving behaviors in a single image. Our work differs from the previous methods in two points. First, we integrate the heatmaps of the driver's head and hands with the color image to extract the pose guided features. Second, we use keypoint action classification to facilitate driving behavior recognition. Besides, we introduce the keypoint gating module to reweigh the keypoint features and extract more discriminative representations. Our method has superior performance on State Farm dataset and R-DA dataset, reaching the state-of-the art level.