10:45 Computer Vision on the GPU – C. Zach (ETHZ)
Using GPUs to accelerate image processing and computer vision tasks is now well established, and predates the now available general processing paradigm of GPUs. Due to the strong connection between computer graphics and computer vision–often both fields together are referred to as “visual computing”–, computer vision researchers soon realized the performance potential of programmable GPUs. We discuss the state of the art of utilizing GPUs for visual 3D reconstruction, and present our contributions to this field. Most steps needed to generate 3D models from images strongly benefit from GPU acceleration: low-level image preprocessing, feature point extraction, tracking and matching, image retrieval, dense multi-view stereo, and high-quality 3D model generation. We’ll illustrate for example how modern GPUs enable the creation of 3D models in real-time from video streams.