Ambarella, Inc. is a leading developer of high-resolution video processing and computer vision semiconductors. The company showcased the CV25 camera System-on-Chip (SoC), which is the latest addition to the CVflow family which combines advanced image processing, CVflow computer vision processing, and high-resolution video encoding in a single, extremely low power consuming design. For extreme home monitoring, the CV25’s CVflow architecture transfers the Deep Neural Network (DNN) processing required for the next generation of intelligent home monitoring, professional surveillance, affordable, and aftermarket automotive solutions. The solution also includes the Driver Monitoring System (DMS) smart dash cameras and electronic mirrors.
The new SoC is aimed to reduce customer’s overall system cost to deliver high-quality image processing, computer vision performance, and advanced cyber-security features at very low power. The features like facial recognition will happen in real-time because the CV25-based cameras perform using Artificial Intelligence (AI). The AI apps in the edge include person recognition and the ability to differentiate between persons, pets, and vehicles. The CV25 can identify familiar faces nearing the front door, decline unknown people from entering, and notify the neighbors when a package is delivered. In DMS, the camera can detect the tiredness by scanning the eyes and facial expressions. The video encoding is efficient in HEVC and AVC formats with low bitrates to reduce the cost of cloud storage. An outstanding image is delivered in low light conditions using the Image Signal Processor (ISP), and maximum image detail can be extracted using High Dynamic Range (HDR). The advanced security suite includes secure boot, I/O virtualization, and TrustZone.
The ultra-low power processing technology is optimized for wire-free camera application which requires long battery life and small form factors. The CV25 chip belongs to the same CVflow family as the existing CV22 and CV2 SoCs, and they share the same computer vision (CV) tools, SDK, ISP, and cybersecurity features. The common factors allow the customers to easily port their neural networks onto CV25. The set includes a debugger, compiler, and support for industry ideal machine learning frameworks like TensorFlow and Caffee, with through guidelines and principles for performance optimizations of convolutional neural networks.