A high-speed driver behavior detection deep learning system using the amount of change in contrast between frames

Abstract

This paper proposes a deep learning system that detects driver behavior for safe driving. A method of detecting the dangerous behavior of an existing driver uses a method of deep learning object detection that detects a class and a location of an object in an image. However, the deep learning object detection algorithm uses many computational resources, so it cannot be used in vehicle embedded environments with limited computational resources. In the case of an object classification algorithm that classifies a single object in an image, fewer computational resources are used than that of a deep learning object detection algorithm. However, it cannot be applied because various objects in the camera image cannot be classified as a single object. In the paper, We propose an algorithm that infers the driver’s behavioral area using the driver’s static movement in a vehicle and then applies deep learning objects to the inferred area. The proposed algorithm may be applied to a vehicle embedded environment because the calculation time is faster and more accurate than the deep learning object detection algorithm.