Evaluation of 3D-Video Compression for Automotive Stereo Vision Systems

Abstract

This paper is an evaluation of the distribution of quantized coefficients resulting from the disparity of an automotive stereo vision system. The system captures the scene with two cameras, computes the disparity with Semi-Global Matching and encodes the left view and the disparity for transmission. Real world and synthetic video sequences were used to evaluate the coefficient distributions of the system under normal and challenging weather conditions. The results show, that the quantized disparity coefficients in frequency space have consistently lower entropy compared to the coefficients of the video scenes. Therefore, it is advantageous for the system to compress the disparity instead of one of the two video streams.