Scaling of Depth Map for Stereo Vision Tasks

Abstract

The article proposes an algorithm for reducing the resolution of reference depth maps (Ground Truth). The evaluating method is presented as well as a comparison of error values resulting from the reference depth map size reduction to various scaling algorithms. The aim of the work is to evaluate the accuracy of scaling depth maps by the proposed algorithm. The need for scaling arises in the tasks of reducing existing different test data sets to uniform parameters of a stereo system, one of which is sensor resolution. A generic data set with the same parameters is required for designing more accurate hybrid stereo vision algorithms with elements of artificial intelligence.