Driver Attention Assistance by Pedestrian/Cyclist Distance Estimation from a Single RGB Image: A CNN-based Semantic Segmentation Approach

Automotive companies are investing a relevant amount of resources for designing autonomous driving systems, driver assistance technologies, and systems for assessing the driver's attention. In this context, two important applications consist in processing images of the surrounding environment to respectively separate the different objects in the scene (semantic segmentation) and to estimate their distances. In both applications, methods based on Deep Learning (DL) and Convolutional Neural Networks (CNN) are being increasingly used, considering LiDAR scans or RGB images. However, LiDAR scanners require dedicated sensors, high costs, and post-processing algorithms to estimate a dense depth map or a three-dimensional representation of the surrounding environment. Moreover, current methods in the literature based on RGB images do not consider the combination of semantic segmentation and depth estimation for assessing the distances of specific objects in the scene. In this paper, we propose the first method in the literature able to estimate the distances of pedestrians/cyclists from the vehicle by using only an RGB image and CNNs, without the need for any LiDAR scanner or any device designed for the three-dimensional reconstruction of the scene. We evaluated our approach on a public dataset of RGB images captured in an automotive scenario, with results confirming the feasibility of the proposed method.