TMVM: Truncated Majority Voting Method to Discriminate and Reduce Mismatches for Local Matching Approaches

Local matching approaches are still common tools in real-time applications. Mismatch is a common situation in stereo vision, especially in local approaches. In this paper, we propose a truncated majority voting method (TMVM) to discriminate and reduce mismatches for local matching approaches in stereo. Experiments on Middlebury benchmark show that mismatches can be discriminated and reduced correctly by deploying the proposed method without losing real-time properties.