Image-based depth estimation in autonomous driving environment

Abstract

Depth estimation is one of the critical problems for autonomous vehicles (AVs) in 3D building modeling, SLAM, and localization. One solution is to leverage Radar and LiDAR to capture sparse point clouds, while the hardware is expensive for wide applications. Alternatively, deep learning (DL) networks have accomplished high performance in various vision tasks including image-based depth estimation, where only a camera is needed for image capturing. In this project, we investigate models to estimate depth using pure image information. We also conduct experiments to demonstrate the effectiveness and to evaluate the performance of these investigated models.