A Fast Local Stereo Matching Algorithm with Simple Edge Detection and Its VLSI Implementation

  • 何 孝威

Student thesis: Master's Thesis


In this thesis a fast local stereo matching algorithm and its VLSI implementation are proposed Unlike the traditional stereo matching methods which process all pixels in the image the proposed algorithm only processed a part of the image to reduce the computation complexity The proposed algorithm generates a sparse depth map by a skip technology For each image this method estimates the skip region in two directions and generates two rough depth maps We compare these two rough depth maps to find the edge of depth map Then re-compute the depth information of these regions with the restricted search range Final we adopt the left right consistency check to fetch the mismatch regions and refine them by neighboring pixels to generate an accurate depth map Therefore the proposed method can reduce at least 74% computation with different settings The error rates of the depth maps are less than 8% The VLSI implementation of the proposed algorithm is also proposed The designed hardware supports the disparity range up to 20 levels and can perform over 30 frames per second each image pair with the resolution of 434 × 383 in 40 MHz
Date of Award2014 Sep 2
Original languageEnglish
SupervisorBin-Da Liu (Supervisor)

Cite this