Cross-comparison census for colour stereo matching applied to intelligent vehicle

摘要

A fast and efficient colour stereo matching algorithm is presented with intended application to intelligent vehicles. The algorithm is initialised by selecting an appropriate colour space giving the smallest disparity error on a training set. A dynamic cross-based aggregation region is then applied. To be fast and robust to noise and illumination variations, a new technique is proposed for census transform in combination with a sparse mask. The algorithm is accelerated with GPU implementation. The performance of the proposed algorithm has been tested on the reference Middlebury dataset, as well as on simulated road traffic scenes of the TNO MARS/Prescan database.