Monocular Depth Estimation Using State-of-the-art Algorithms: a Review

Monocular Depth Estimation is the process of calculating the depth value of each pixel given a single RGB image. This challenging computer vision task is the main prerequisite for determining scene understanding for applications such as 3D scene reconstruction, augmented and virtual reality. Additionally, a lot of robotics issues, like mapping, localization, and obstacle avoidance for terrestrial and aerial vehicles, depend on depth information. Five monocular depth estimation techniques are compared. This comparison focuses on how generalizable the methods are. According to this study, monocular depth estimation techniques frequently exhibit artifacts when used on images that are not part of the training set, despite performing well on images that are similar to the training images. We test the various approaches using photos that resemble training data as well as paintings or images with odd perspectives.