Arm has launched a new image signal processor (ISP), the Mali-C55, which it claims is its smallest yet — but that delivers higher image quality and boosted performance over its predecessor, and is built with multi-camera and Internet of Things (IoT) computer vision systems firmly in mind.
"The Mali-C55 delivers upgraded image quality features, works under a wide range of different lighting and weather conditions," claims Mohamed Awad, Arm's vice president for IoT and embedded, "and is designed to enable maximum performance and capability in area and power constrained applications, making it ideal for smart camera and edge AI vision use cases."
What use cases? The company claims that the part can boost detail in surveillance and security applications, including being able to pick out a license plate on a vehicle traveling at up to 75 miles per hour, offer higher-resolution footage from connected security cameras, and integrate secure visual unlock into smart home hubs.
The Mali-C55 ISP itself offers eight multi-sensor camera inputs with image resolutions up to 8K, a claimed 1.2 gigapixel per second throughput, and a maximum image size of 48 megapixels — which can be increased by adding extra Mali-C55s — doubling the number of inputs and throughput and tripling the maximum resolution over the company's earlier Mali-C52 ISP. To boost quality, the processor includes better tone mapping and spatial noise reduction, enhanced support for high dynamic range (HDR) sensors, and support for external machine learning accelerators for neural network operations including advanced de-noising on top of the part's on-board Temper temporal and Sinter spatial noise reduction engines.
"As ML moves closer to the edge, advanced image processing can be leveraged by integrating more ISPs into the SoCs [Systems-on-Chip]," says Awad. "By enabling easy integration between Mali-C55 and machine learning accelerators, we’re delivering new levels of on-device processing in devices that require high quality vision systems because the output from the ISP can be sent directly to the ML accelerator. This reduces cost and processing time by having less data sent from device to cloud, without having to compromise on inferencing."
More information on the new part is available on the Arm Developer site; the company has also pledged to bring the ISP into the Arm Total Solutions for IoT program in the near future. An additional technical blog, penned by Arm's Parag Beeraka, dives into more detail.