Adaptive cost aggregation table on conditional random fields for intelligent vehicles

Many stereo vision algorithms for intelligent vehicles use a cost aggregation strategy because of its efficiency. We propose Adaptive Cost Aggregation Table (ACAT), an algorithm that is a global cost aggregation method which uses every cost in the whole image to estimate each disparity. The proposed algorithm works on conditional random fields to use locally variant information. ACAT aggregates cost adaptively, considering local intensity information. However, computational complexity does not increase compared to standard Semi-Global Matching (SGM) and Cost Aggregation Table (CAT) algorithm. We compared the proposed algorithm to other cost aggregation algorithms in analysis of the KITTI dataset. Disparity results of ACAT were more accurate than those of SGM and CAT, for f at areas, for discontinuous areas and in occlusion areas.