[ShangHai, China, March 15, 2021] Artificial intelligence (AI) has transformed whole industries, and HiSilicon has put the promise of AI into practice, thanks to its state-of-the-art capabilities in AI computing power, algorithm tools, frameworks, and application enablement.
The Baidu Dumu CM-Mini Panel, which incorporates HiSilicon's AI SoC, has recently hit the market. It offers stellar facial recognition performance in bright and low-light conditions alike, and for all age groups, by virtue of a whopping 50,000 facial image database. Facial recognition can be completed within 300 ms, and is available for both cloud and on-device services.
CM-Mini: Compact and Comprehensive
The CM-Mini has it all: a 5-inch touchscreen, metallic frame, and dual near-infrared lens (capable of fast and stable facial recognition at all hours), with default support for a microphone, speaker, card swiping, light sensor, pulse notification light, and tamper-proof switch.
HiSilicon AI: The Fruit of Relentless Innovation
HiSilicon has been working towards providing basic AI computing platforms for security protection. Since years ago, HiSilicon has launched a high-end smart vision processing chipset that has drawn acclaim, and packs a multitude of features, including: 8K30/4K120 broadcast-level digital video recording, multi-channel sensor input, H.265 output or cinema-grade RAW data output, high-performance image signal processor (ISP), and industry-leading low-power technology with a power-efficient architecture.
The CM-Mini uses the next-generation industry-dedicated Smart HD IP camera SoC, which integrates the new-generation ISP, the latest H.265 video encoder, as well as a high-performing neural network inference engine (NNIE). This SoC stands out from the crowd, due to its low bit rate, high image quality, intelligent processing and analysis, and low power consumption. Since it comes integrated with the POR, RTC, audio codec, and standby wakeup circuit, the SoC is capable of dramatically reducing EBOM costs for customers. Its interfaces are similar to those in HiSilicon DVR and NVR chipsets, which is conducive for seamless and efficient product development as well as mass production.
Sees in the Dark
Facial detection and recognition performance on the device is consistently robust, even in poor lighting conditions, featuring backlight, strong light, or even little to no light. This means that users can benefit from all-purpose facial recognition, which requires no extra effort on their part.
Works for All Ages
The facial recognition supports users aged from 6 to 90. Thanks to dynamic learning, the device enhances face recognition accuracy and speed over time. Even when deployed in areas frequently occupied by children and elderly people, the device still manages to deliver next-level performance.
Finance-level Liveness Detection
Only Real Activities Make it Through
Binocular multimodal liveness detection, including the synthetic image defense algorithm, prevents spoofing attacks on a comprehensive basis. Tests have revealed that the device can withstand 99.9% of spoofing attacks, including photo attacks, video attacks, and 3D mask attacks.
Access at a Glance
Full-process searching requires less than 300 ms when there are 50,000 face images in the face database. Due to the in-depth optimization of access policies, faces can be recognized so long as certain conditions are met, even when the user is running or cycling.
Broad-Ranging Application Scenarios
● Enterprise attendance: Clock-in/out at the reception desk and sign-in/out for meeting rooms
● Smart community: Visitor management, room access verification, and facial recognition-based staircase controls
● Smart campus: Dormitory management, teaching building access verification, and exam verification
● Building access control: Visitor registration, access control, and stranger management
● Smart zone: Visitor management, payment, card-free access, and access control management