Using the GPU for fast symmetry-based dense stereo matching in high resolution images

摘要: SymStereo is a new algorithm used for stereo estimation. Instead of measuring photo-similarity, it proposes novel cost functions that measure symmetry evaluating the likelihood two pixels being match. In this work we propose parallel approach LogN matching variant capable processing pairs images in real-time depth The power graphics units utilized allows exploring more efficiently bank log-Gabor wavelets developed to analyze symmetry, spectral domain. We tradeoffs and different parameter-izations signal accommodate image size, dimension filter bank, number also disparities controls space density estimation, still process up 53 frames per second (fps) with size 288 × 384 3 fps 768 1024 images.