Low-power, Secured, Learn-on-the-fly Facial Recognition Multi-Platform Module

Facial recognition is one of the most popular non-invasive security technologies in use, given the ever-present threats in both the physical and cyber world. Although facial recognition systems has seen tremendous growth, the technology has been hampered by cost and energy consumption limitations. Real-time training and inference of faces require huge computing power, typically using high-end and power-hungry CPU (central processing unit), GPUs (graphics processing units) or SoCs (system on a chip). The threat of cyber-hacking the stored facial bio-data is also real, as these digital data remain housed in IT server infrastructure.

This technology is based on a hardware implementation of Artificial Neural Networks, offering a low power, low cost and secure multi platform module. No software is required, and the architecture allows dynamic learning-on-the-fly.

• This technology implements ANN architecture in IC or FPGA.
• Identical neuron cells combining memory and processing using Radial Basis Function (RBF) technique.
• The technology is content reactive, and no addressing is required. All neurons are interconnected and work in parallel.
• The low power on-the-edge implementation and interfacing is designed to cater for a variety of microcontrollers.
• Real-time facial features learning and recognition.
• Deterministic latencies to learn and recognize faces regardless of the number faces registered.
• Dynamic learning, in which real-time addition of new faces (ability to learn on-the-fly and requires less training datasets).

• Surveillance Cameras
• Access control
• Security Robots
• etc.

• Low hardware cost
• Low power requirements (in milliwatts)
• Fast inference speed (in milliseconds)
• Real-time facial features learning and recognition
• Deterministic latencies to learn and recognize faces regardless of the number of faces registered
• The system provides real-time addition of new faces, i.e. able to learn-on-the-fly, hence requiring less training datasets
• Scalability for increasing facial database