Monocular Depth Estimation Using Deep Learning: A Review

Abstract

In current decades, significant advancements in robotics engineering and autonomous vehicles have improved the requirement for precise depth measurements. Depth estimation (DE) is a traditional task in computer vision that can be appropriately predicted by applying numerous procedures. This task is vital in disparate applications such as augmented reality and target tracking. Conventional monocular DE (MDE) procedures are based on depth cues for depth prediction. Various deep learning techniques have demonstrated their potential applications in managing and supporting the traditional ill-posed problem. The principal purpose of this paper is to represent a state-of-the-art review of the current developments in MDE based on deep learning techniques. For this goal, this paper tries to highlight the critical points of the state-of-the-art works on MDE from disparate aspects. These aspects include input data shapes and training manners such as supervised, semi-supervised, and unsupervised learning approaches in combination with applying different datasets and evaluation indicators. At last, limitations regarding the accuracy of the DL-based MDE models, computational time requirements, real-time inference, transferability, input images shape and domain adaptation, and generalization are discussed to open new directions for future research.

Parallax Inference for Robust Temporal Monocular Depth Estimation in Unstructured Environments

“…Recent surveys [ 9 , 10 , 11 , 12 ] present summarized descriptions and comparisons of single image depth estimation methods. The main observation made in these surveys is that estimating depth from a single image remains difficult, especially for autonomous vehicle applications.…”

“…As in previous works [50][51][52], we use a multi-scale loss function. For each frame and each level, we compute the L 1 distance on the logarithm of the depths resulting from the conversion of parallax estimates using Equation (9). The logarithm leads to a scale invariant loss function [43] and the use of an L 1 distance is motivated by its good convergence properties [53].…”

“…However, datasets suitable to study the task of depth estimation in environments with little structure—we call them “unstructured environments” in the following—were proposed only recently (see Fonder and Van Droogenbroeck [ 6 ], Wang et al [ 7 ]). As a result, existing methods for outdoor depth estimation [ 8 , 9 , 10 , 11 , 12 ] are benchmarked against older datasets that only target autonomous driving applications in urban environments.…”

Nested DWT–Based CNN Architecture for Monocular Depth Estimation

“…The network model architecture consists of customized building blocks, such as convolution layers, pooling functions, activation layers, and expansion layers [ 6 , 7 , 8 ]. The state–of–the–art CNN use supervised or self–supervised training strategies [ 6 , 9 , 10 , 11 , 12 , 13 ]. Supervised training [ 5 ] requires a labeled depth image for the network to converge.…”

Maritime Tracking With Georeferenced Multi-Camera Fusion

“…Binocular depth estimation, also known as stereo vision, is an alternate approach that relies on triangulating two cameras with overlapping fields of view. Pixels in each image are matched to each other using a wide range of methods from classical [14] to deep learning-based [15]. These methods are less reliant on large training data but can still be computationally expensive and large baselines are typically required for most distances encountered in a maritime context.…”