Sparse Area Stereo Matching Experiment

摘要

The algorithm used in this experiment is considered by many to be state-of-the-art for calculating x-parrallax over rural regions. Even so, its output must be refined in sparse areas in order to meet accuracy requirements. The major result of this experiment showed that x-poarallax should be measured from digital imagery containing no more than 10 lone pairs per millimeter. The objective of this report is to find ways to improve the output of the X-parallax front-end processor without using information from other front-end processors, and without drastically revising the algorithm. Two approaches were tested, namely to provide to the processor image data most suitable for rural type matching and to use similar yet less demanding procedures in difficult areas. In the first case, stereo matching was performed with digital pictures containing up to 10 lp/mm of information, and in the second case large correlation windows were used on difficult points. The latter approach was performed with another DIAL program designed to measure corresponding points in a pass point mode.