London-based Plumerai has ported its people detection neural network algorithm SimpleLink CC3220SF wireless microcontroller and PIR sensor development from TI.
A PIR sensor is typically able to detect large motion, for example someone entering a room or moving around, but not smaller ones. The Plumerai binarised neural network allows the microcontroller, based on an ARM Cortex-M4 running at 80MHz, to pick up small movements with accuracy of 88 percent. The model also does not focus on one person, minimising false triggers. The small model size of 154kbytes also fits in the 1Mbyte of Flash and the peak RAM usage is 170Kbytes to fit into the 256Kbytes of SRAM to fit alongside the WiFi protocol stack. This is 100 times smaller than an equivalent GPU-based neural network with power under 100mW for less than $10.
“It’s great to be working with TI and to enable many more AI applications on low power and small MCUs. Together we boldly go where no AI has gone before,” said Marco Jacobs, head of product marketing at London-based Plumerai
- Plumerai develops embedded AI accelerator IP core for FPGAs
- World’s fastest deep learning inference software for ARM cores
- XMOS and Plumerai partner on binarised neural networks
Other articles on eeNews Europe
- e-ink display for adaptive car exteriors
- Melexis teams for anti-spoofing AI with ToF sensor
- Omnivision opens Belgian R&D centre
- e-peas uses new microcontroller in battery free surveillance system
- Infineon partners for ultra-low power UWB chip