Скачать или смотреть Jaccard distance as similarity measure for disparity map estimation

In this manuscript, we report the results of a different similarity measure for sparse disparity map (DM) estimation. The proposed framework includes the proposal for a custom Census transform as a feature extraction algorithm, then a sparse disparity map is estimated with Jaccard distance as a similarity measure, and finally, a consistency match check is performed to label occlusions and invalid estimations. Bad Matching Pixels and Invalid pixels percentage are applied in order to compare the performance of the proposed approach against state-of-the-art techniques. The current approach is designed in order to obtain high-confidence DM estimation with low-cost hardware and no previous data for learning, in comparison, our approach is capable to obtain competitive results with low computational cost