PYRAMID BINOCULAR DEPTH ESTIMATION MODEL WITH SELF-IMPROVING CAPACITY

摘要

A pyramid binocular depth estimation model with a self-improving capacity, relating to the field of deep learning and two-dimensional image depth reconstruction. The following steps are comprised: 1) constructing a space pyramid module; 2) performing cost integration; 3) performing multilayer regression outputting; and 4) calculating a loss function. On the basis of a pyramid binocular model, in the model, a binocular image is reconstructed by using an SPN (space transformation network) according to a parallax image generated by the model, and the model is trained according to the loss between the reconstructed image and an original image, so that the model can be trained without any preprocessed radar data; and when there is preprocessed radar data, the supervised and trained model can be trained online by means of binocular reconstruction loss during a test so as to improve the performance thereof.