Real-time Stereo with Dense Output by a SIMD-computed Dynamic Programming Algorithm

We present a modified stereo algorithm based on dynamic programming techniques, which outputs a dense disparity map, and its actual implementation on a linear SIMD image processing device with 256 PE, the IMAP-VISION. Our results show the high suitability of this algorithm and the parallel hardware for real-time 3D-vision based robot control. Running on the SIMD device, the algorithm guarantees a processing time of less than 100 ms hard for a Region Of Interest of 128 by 128 pixel and a respectively large depth range of 64 disparity levels. The algorithm can be adjusted with ease and fast at the update rate of the input for even higher output rates, when considering lower depth resolution or less ROI width.

[1]  Jin-Jang Leou,et al.  A dynamic programming approach to line segment matching in stereo vision , 1994, Pattern Recognit..

[2]  Hong Jeong,et al.  Fast Stereo Matching Using Constraints in Discrete Space , 2000 .

[3]  Takeo Kanade,et al.  Stereo by Intra- and Inter-Scanline Search Using Dynamic Programming , 1985, IEEE Transactions on Pattern Analysis and Machine Intelligence.

[4]  Carlo Tomasi,et al.  Depth Discontinuities by Pixel-to-Pixel Stereo , 1999, International Journal of Computer Vision.

[5]  Ingemar J. Cox,et al.  A Maximum Likelihood Stereo Algorithm , 1996, Comput. Vis. Image Underst..

[6]  Gruia-Catalin Roman,et al.  A parallel algorithm for incremental stereo matching on SIMD machines , 1990, [1990] Proceedings. 10th International Conference on Pattern Recognition.

[7]  Phil-Woo Han,et al.  A new stereo matching hardware with merged odd-even PE architecture , 1999, Proceedings of IEEE. IEEE Region 10 Conference. TENCON 99. 'Multimedia Technology for Asia-Pacific Information Infrastructure' (Cat. No.99CH37030).