Improved Real-Time Stereo on Commodity Graphics Hardware

This paper presents a detailed description of an advanced real-time correlation-based stereo algorithm running completely on the graphics processing unit (GPU). This is important since it allows to free up the main processor for other tasks including high-level interpretation of the stereo results. Compared to previous GPU-based stereo implementations our implementation includes some advanced features such as adaptive windows and cross-checking. By taking advantage of advanced features of recent GPUs the proposed algorithm is also a lot faster than previous implementations. Our implementation running on an ATI Radeon 9800 graphics card achieves over 289 million disparity evaluations per second including all the overhead to download images and read-back the disparity map, which is several times faster than commercially available CPU-based implementations.