The project is aimed to create and deploy a smart doorbell application that would be highly energy-efficient. The main application pipeline consists of 4 stages: frame capture, face detection, face recognition and user interaction e.g. displaying message. In addition to regular face detection and re-identification loop, application provides functionality to add known people to the trusted list using already calculated face descriptors.
The demo is targeted for two boards Gapoc A v2 and Gapuino v2 and GAP SDK release 3.6 and 3.7.1. GAP8 V1 chips are not supported in SDK any more. List of extra components is provided below.
For Gapuino board:
- HIGHMAX camera module
- Adafruit 2.8 TFT display with SPI interface
For Gapoc A board:
- Adafruit 2.8 TFT display with SPI interface
- GAPOC_A Adapter for Adafruit LCD 2.8 version 3 or version 4 or hand-made shield with push button
- Android-based smartphone with pre-built user management application
- Hardware configuration and schematics
- Build and test instructions
- Pipeline overview
- ReID train instruction
- ReID network quantization for GAP
- ReID network architecture and inference details
- Bluetooth LE protocol for users management
- 512KiB RAM Is Enough! Live Camera Face Recognition DNN on MCU: link
@InProceedings{Zemlyanikin_2019_ICCV,
author = {Zemlyanikin, Maxim and Smorkalov, Alexander and Khanova, Tatiana and Petrovicheva, Anna and Serebryakov, Grigory},
title = {512KiB RAM Is Enough! Live Camera Face Recognition DNN on MCU},
booktitle = {The IEEE International Conference on Computer Vision (ICCV) Workshops},
month = {Oct},
year = {2019}
}
- GreenWaves press release and demonstration video: link