Dec 14 2012
Vivante Corporation, a world-wide leader in graphics and compute technologies, today announced its cores have been widely adopted to power the next generation ADAS Computer Vision applications from top tier automotive OEMs. ADAS is a term that describes the latest electronic technologies found in automobiles that focus on increasing safety for occupants, pedestrians, and surrounding vehicles.
These safety technologies use sensor fusion to combine inputs from cameras, sensors, GPS, RADAR, and IR to create "intelligent" cars that are very aware of their surroundings. These real time systems continuously monitor, predict, and try to prevent accidents. Automobiles from Acura, BMW, Ford, GM, Mercedes, and Volvo have safety packages that include different forms of ADAS.
What the automotive industry is saying about ADAS and OpenCL™
"We definitely see OpenCL gaining momentum as the industry emphasizes safety, features, and convenience," said Dan Loop , Freescale Semiconductor i.MX Automotive Product Marketing. "One reason for this rapid adoption of new safety features is the traction OpenCL is getting from OEMs and developers. The programmability and speed-up of some critical computer vision algorithms on the GPGPU provide significant performance gains and power savings over DSPs and CPUs. Using the OpenCL solution available in the i.MX 6 platform, our customers optimize their algorithms for hybrid systems to take full advantage of the GPU as well as the CPU."
In a publication from research firm ABI Research, they report "this year [2011], global ADAS market value should approach $10 billion, growing to an impressive $130 billion in 2016."1
Two reports from Research and Markets about ADAS state "recent growth in installations of ADAS is set to accelerate and provide opportunities for those innovative suppliers participating."2 A second report focusing on the EU market mentions "Moving ADAS from niche to mass market: Europe is witnessing a surge in the availability of ADAS, reflecting the joint efforts of the automotive industry and European governments to improve safety and enhance driver comfort."3
Another industry report4 concludes that factors supporting ADAS are "Inclusion of ADAS in new car safety assessments … by Euro NCAP and NHTSA", "EC legislation requiring fitment of ADAS in trucks and buses will come into force at the end of 2012 and private cars are expected to follow", and "vehicle manufacturers are continuing to reduce system prices and migrate ADAS availability to mass market models."
The Vivante ADAS OpenCL™ Solution
Vivante's second generation fully programmable ScalarMorphic™ architecture processes thousands of concurrent threads to achieve gigaflops (GFLOPS) of high precision, real-time computational performance. OpenCL ADAS calculations leverage parallel GPU SIMD engines to boost computation density by performing massively parallel data processing on multiple data streams simultaneously. Vivante GPGPU solutions are built around the just released OpenCL 1.2 Full Profile specification that enables refinement of the most important parallel algorithms.
For certain classes of problems, the GPU is far more computationally and power efficient than other SIMD vector processors with a similar number of functional units. The GPGPU paradigm shines on "embarrassingly data parallel" problems, which are ubiquitous in Computer Vision applications like ADAS. More specifically, computations requiring heavy use of floating-point arithmetic and/or image warping and resampling will see significant benefits.
As part of the heterogeneous ADAS architecture in the Freescale i.MX 6 automotive grade applications processor, Vivante cores process incoming data streams from cameras and other sensors to accelerate the parallel portions of CV algorithms. Typical ADAS computations that see significant speed-up on the GPU include SURF, SIFT, Hough Transform, Canny or Sobel edge detectors, HOGS (Histogram of Oriented Gradients), Integral Image, and point cloud processing. These algorithms can also be used in other areas ranging from Eye Tracking, Face and Gesture Recognition Natural User Interfaces, classical Computer Vision, and advanced image processing for CE, Industrial, and Mil-Aero products.
"OpenCL on the GPU has now become a fundamental part of the latest heterogeneous compute architectures. ADAS is one key area where the jump from GPU research exercise to real world implementation has been made. We are proud to partner with Freescale's automotive team to bring leading safety and computer vision features to the next generation of cars," said Wei-Jin Dai , President and CEO of Vivante.