Computer Vision Meets Visual Computing : Augmented Reality & The GPU Trak Lord, US Marketing & Media Relations
http://www.facebook.com/metaio @twitt_AR http://twitter.com/#!/twitt_AR http://augmentedblog.wordpress.com/ feedback encouraged @traklord http://www.flickr.com/photos/metaio/ http://www.youtube.com/user/metaioAR @metaioUS #SIGGRAPH Phone (EMEA): +49-89-5480-198-0 info@metaio.com Phone (US): +1-415-814-3376 www.metaio.com
Something to talk about • Company background • AR Graphics pipeline • The Demand for AR • AR for hardware • NVIDIA • Conclusion Q&A
Who is Metaio? • Leading Augmented Reality Software • 500+ B2B customers (IKEA, Lego, & Solutions Provider Audi, Mitsubishi Electric) • Extensive Patent & IP Portfolio • Munich HQ + San Francisco Office • Decade of AR • Only AR Company to offer solutions, middleware, hardware IP & content • AR SDK with 50k Developers • Over 300 AR apps powered by Metaio
Who is Metaio? http://www.youtube.com/watch?v=-24ySXEMq1U
How can AR help Graphics? The Dream: A graphic designer can take any 3-D asset and instantly transfer it to the real world, where he or she can instantly do two things: • Visualize it in its natural environment • Compare it to its real-world counterpart
Augmented Graphics • More precise models • Simplified content creation pipeline • Bone animations • Motion builder • .FBX support
Augmented Graphics http://www.youtube.com/watch?v=_Ij9kcWsni4
Augmented Graphics http://www.youtube.com/watch?v=DKEcQ9uiII0
The Demand for Augmented Reality Things we already know • Augmented Reality is a useful technology • It provides value to companies all over the world • Visualization technology can reduce need for physical materials • 2-D has limits
The Demand for Augmented Reality +22M Views +20M Views +330K Views +4M Views
The Demand for Augmented Reality The software is there, the algorithms are there, the capabilities are there, but the hardware just isn’t.
The AREngine: A hardware solution for AR http://www.youtube.com/watch?v=6br7NreTwD4
More Compute Power Is Good For AR CPU Compute Tasks (software level) • Descriptors matching • Feature extraction • Algorithms • Camera • Display • Cloud connectivity
More Compute Power Is Good For AR
More Compute Power Is Good For AR
More Compute Power Is Good For AR
3-D Sensor Pipeline Sensor Image Capture Capture Pre-Processing / Color Feature Conversion Extraction Algorithmic hot spots Building Feature Pose Estimation Feature Matching Descriptors Rendering 3D Objects Image Composition Capture Image Upload
More Compute Power Is Good For AR • Far more expensive to MOVE data than COMPUTE data Write 32-bits to Memory 600pJ • Energy efficiency is important for both silicon AND software design • Core stays idle when HW IP is ON Send 32-bits Off-chip • Local SRAM for power efficient zero latency 50pJ access • Need to use hardware acceleration • Feature Matching by AREngine gets up to 60X faster than doing it on core Send 32-bits 2mm • Efficient caching and memory usage 24pJ • Reduces data movement • A must for comparing large databases • Faster object recognition and location 32-bit Float Operation For 40nm, 1V process • Smoother user experience 7pJ • No longer limited to 2-D planar tracking 32-bit Integer Add 1pJ 32-bit Register Write 0.5pJ Energy spent during typical MOVE/COMPUTE operations (Source: Khronos Group/NVIDIA)
More Compute Power Is Good For AR 20
More Compute Power Is Good For AR Problems • power consumption • feature matching (most expensive in memory matching and energy) • data and end user experience varies, but feature matching is always the same process
More Compute Power Is Good For AR Solutions • GPU / CPU Optimization • OpenCL (software) • CUDA • Dedicated Hardware IP
More Compute Power Is Good For AR Results • Feature Matching by AREngine gets up to 60X faster than doing it on core • Faster object recognition and location • Smoother user experience • No longer limited to 2-D planar tracking • Reality becomes a canvas, upon which we can design the virtual intersection of the two worlds
Metaio SDK (metaio.com/products/sdk) • Optimized client-based visual search for +100 images • Improved stability, robustness and performance of ID Marker and 2-D markerless tracking • Improved performance of 3-D object tracking and training • SLAM integration • Sensor fusion for visual-based tracking • AREL (HTML5 & JavaScript) http://www.youtube.com/watch?v=EltxfrbDZIg
Join us at this year’s InsideAR metaio.com/insideAR http://www.youtube.com/watch?v=SvU1pDANDkU
thank you + Q&A @metaioUS info@metaio.com http://bit.ly/metaioblog http://youtube.com/MetaioAR
Recommend
More recommend
