Ultralow-Power AI Visual Sensor for Battery Operated & IoT Applications

This technology offer is an always-on artificial intelligence (AI) visual sensor that consumes less than 1 mW of power, and is suitable for use in next-generation, battery-operated smart camera systems. The design combines an ultralow-power image sensor with a custom computer vision application-specific integrated circuit (ASIC). It runs optimised machine-learning (ML) algorithms on the edge, without the need to stream images to the cloud. This allows for quicker response times and privacy protection.

IoT System-on-Chip (SOC):

• Extremely efficient ML algorithm
• Human detection, face detection and recognition, people counting, etc.
• Efficient processing of low-resolution images
• Less than 2 MB memory footprint
• Less than 1mW always-on mode
• Cost/power efficiency
• Computer vision hardware accelerators

 CMOS Sensors:

• Less than 1 mW always-on monitoring
• qVGA/VGA for IoT
• HD RGB/IR for Notebooks

• Security: AI enhancement of smart cameras
• Smart Home: Human sensors and smart motion detection
• Notebook: Always-on sensors of human face/presence
• Smart Buildings: Occupancy and people counting
• Consumer Appliances: Human presence
• Automotive: Access control and security
• Industrial: Computer vision sensors
• Elderly: Fall and imbalance detection

The system’s ultralow-power design, unique image processing pipeline and hardware accelerators allows battery-operated, AI image processing at the edge. As such, a wide range of design possibilities and applications are enabled.