On deep learning techniques to boost monocular depth estimation for autonomous navigation

Raul de Queiroz Mendes, Eduardo Godinho Ribeiro, Nicolas dos Santos Rosa, Valdir Grassi Jr.

Abstract: Highlights•Fast and lightweight convolutional network to address monocular depth estimation.•Novel feature extraction models applicable to several Deep Learning architectures.•Simple and efficient surface normals module, with a new 2.5D geometric loss function.•Comprehensive surveys of the single image depth estimation and depth completion areas.•Extensive and exhaustive ablation studies on different Deep Learning strategies.