Monocular Supervised Metric Distance Estimation for Autonomous Driving Applications

Abstract

The task of monocular distance estimation is a major area of research in the computer vision field. Especially relevant this task is to the autonomous driving applications, where robustness and accuracy of the distance estimation significantly affect driving safety. In this paper we propose a simple, fast and efficient deep learning model capable of extracting distance information for a detected object from monocular images. The model is trained and tested on the KITTI benchmark and compared to the Monodepth2 model. The conducted experiments show that the proposed convolutional neural network architecture outperforms Monodepth2 by 11% on average according to the weighted average mean absolute error.