Enhancing Self-Supervised Depth Estimation Through Camera Parameter Priors

Jinchang Zhang, Xue Iuan Wong, Guoyu Lu

Keywords: Depth Estimation, Camera

Abstract:

Depth estimation is a key topic in the field of computer vision. Self-supervised monocular depth estimation offers a powerful method to extract 3D scene information from a single camera image, allowing training on arbitrary image sequences without the need for depth labels. However, monocular unsupervised depth estimation still cannot address the issue of scale and often requires ground-truth depth data for calibration.
In the deep learning era, existing methods primarily rely on relationships between images to train unsupervised neural networks, often overlooking the foundational information provided by the camera itself. In fact, based on physical principles, the camera’s intrinsic and extrinsic parameters can be used to calculate depth information for the ground and related areas and extend it from planar regions to full scene depth. To make full use of scene depth, even in the presence of errors, we introduce a contrastive learning self-supervised framework. This framework consists of two networks with the same structure: the Anchor network and the Target network. The predictions from the Anchor network are used as pseudo-labels for training the Target network. Depth reliability is determined by entropy, dividing the predicted depth into positive and negative samples to maximize the use of physical depth information, and effectively enhance the depth estimation accuracy.

Primary Area: applications to computer vision, audio, language, and other modalities

Submission Number: 3621