The adoption of 4K video is exploding, and drones, robots, cars and virtual reality headsets are set to join the party.
Chip companies are at CES showing off their efforts to ease 4K video playback and capture on consumer electronics. And in addition to supporting the boosted video resolution, new chips and architectures are using artificial intelligence features to recognize and classify images.
Ambarella announced H2 and H12 low-power chips for small 4K cameras found in drones; it also has low-power ARM Cortex CPUs for flight control and wireless data transfers. The H2 and H12 have image stabilization and HDR (high-dynamic range) features for sharper and brighter images. The chips support AVC and H.265 4K video formats.
Graphics company Imagination Technologies has a new graphics architecture that can recognize objects in 4K images. Beyond smartphones and tablets, the new PowerVR Series7XT Plus family of mobile graphics processors is targeted at virtual reality headsets, robots, drones and cars.
The new GT7200 Plus and GT7400 Plus GPUs have specific computer vision instructions to recognize images, which could help drones, robots or cars navigate or execute tasks. Imagination spokesman Alexandru Voica gave some examples: self-driving cars with the GPUs could be more efficient at processing algorithms for lane detection or collision avoidance. A surveillance camera may be able to recognize a thief from a 4K video. Images could also be sent to deep-learning systems in the cloud for identification.
Imagination's new processors could also be useful for on-device graphics processing in standalone 3D virtual reality headsets. But that won't match the experience provided by VR headsets connected to rigs with more powerful graphics CPUs.
In-car entertainment systems will get 4K video with the help of Qualcomm's latest Snapdragon 820A and 820AM processors, which have an Adreno 530 GPU. Much like Imagination's processors, the chips will recognize and analyze objects so cars can respond to on-road situations.
Nvidia announced the Pascal GPU, which is in the Drive PX 2 computer for self-driving cars. The lunchbox-sized supercomputer, shipping to partners in the second quarter, sits in the trunk and provides the horsepower equivalent of 150 MacBook Pros, according to Nvidia CEO Jen-Hsun Huang. The Pascal GPU can help cars recognize objects, traffic lights, speed limit signs and crosswalks. Nvidia's goal is to add more deep-learning instructions to the chip so self-driving cars can, for example, know to pull over by distinguishing an ambulance from other trucks, or recognize snowy conditions and know that lanes are hidden on the road.