vcc_gnd
/
esp-who
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
官方ESP-WHO资料，这里可以找到板子的测试DEMO,推荐乌班图下编译生成固件
5 Commits
6 Branches
10 Tags
30 MiB
C 72.5%
HTML 21.9%
C++ 4.5%
Shell 0.8%
CMake 0.2%
 
 
 
 
 
Go to file
XiaochaoGONG 20eabaac3f add changelog 2018-11-03 11:52:39 +08:00
components update LICENSE 2018-10-29 10:56:16 +08:00
esp-idf@129d32772e Init commit 2018-10-26 21:09:52 +08:00
examples/single_chip/camera_with_command_line update LICENSE 2018-10-29 10:56:16 +08:00
img change overview img 2018-11-02 19:06:22 +08:00
.gitignore Init commit 2018-10-26 21:09:52 +08:00
.gitlab-ci.yml Init commit 2018-10-26 21:09:52 +08:00
.gitmodules Init commit 2018-10-26 21:09:52 +08:00
CHANGELOG.md add changelog 2018-11-03 11:52:39 +08:00
LICENSE Init commit 2018-10-26 21:09:52 +08:00
README.md change overview img 2018-11-02 19:06:22 +08:00

README.md

ESP-WHO

ESP-WHO is a face detection and recognition platform that is currently based on Espressif Systems' ESP32 chip. ESP-WHO is within Espressif Systems AI framework, with the focus on face detection and recognition so far.

Overview

ESP-WHO supports development of face detection and recognition applications based around Espressif Systems' ESP32 chip in the most convenient way. With ESP-WHO, you can easily build up face detection- and recognition-featured applications, for instance:

  • A coffee machine that brews coffee according to your taste preference via ESP-WHO;
  • Home applicance that will shut off the electricity automatically if ESP-WHO recognizes that unsupervised children are operating them;
  • And many more applications that suit your needs.

In general, the ESP-WHO features will be supported as shown below:

overview

In ESP-WHO, Detection, Recognition and Image Utility are at the core of the platform.

  • Image Utility offers the fundamental image processing API for ESP32.

  • Detection, or MTMN, uses MTCNN and MobileNet for reference. However, we replace CNN with MobileNet block and the model size is quite smaller than other face detection models. With special training schemes, MTMN could achieve good performance in detection in some applications scenarios.

  • Recognition is implemented with MobileFace and will be released soon, accompanied with technique details.

Both input and output are flexible.

  • Image sources could be input via camera, Wi-Fi socket and etc. However, we don't provide many drivers right now, but drivers for other camera modules, Wi-Fi socket and etc will be released very soon.

  • Results could be output and displayed on Command line, LCD or even website via Wi-Fi http_server.

Quick Start with ESP-WHO

Hardware Preparation

To run ESP-WHO, you need to have a development board that integrates a ESP32 module that has sufficient GPIO pins and more than 4 MB external SPI RAM. In this case, we take ESP-WROVER-KIT as the test board, which is highly recommended for getting started with ESP-WHO.

On how to select ESP32 module for your applications, please refer to the README.md of each example.

Software Preparation

Image

The recommended resolution of input image is 320x240 in RGB565 format.

As for choosing camera as an image offer, make sure that the ESP32 module you choose offers specific pins your camera needs. If there are pin conflicts, please refer to the pin connection in README.md of each example.

By now, we have provided the driver of OV2640, which is highly recommended as an image offer to get started with. More camera drivers will be released later on.

ESP-IDF

See setup guides for detailed instructions to set up the ESP-IDF:

In this case, we take ESP-IDF v3.1 as the test version.

ESP-WHO

Make sure you clone the project recursive:

git clone --recursive https://github.com/espressif/esp-who.git

If you clone project without --recursive flag, please go to the esp-who directory and run command git submodule update --init before doing anything.

Components

Components such as camera drivers, image utility, face detection and etc are stored in components folder.

See the following sections for more details about components.

Camera

The camera component contains drivers for camera devices of ESP32. Now OV2640 is supported.

Image_util

The image_util component is designed for fundamental image processing of ESP32. Resizing and transforming are the major operations in ESP-WHO.

esp_facenet

The esp_facenet component contains the APIs of ESP-WHO neural networks. Only face detection is released by now.

  • esp_facenet/mtmn: The neural network of MTMN is stored in this folder. Codes in mtmn.c file provide detailed implementations of P-Net, R-Net and O-Net and offer the APIs.
  • esp_facenet/face_detection: The face detection implementation is stored in this folder. Codes in face_detection.c call net implementation APIs of mtmn and also with some process such as nms to implement face detection.

Face Recognition of ESP-WHO will be stored in here as well.

Examples

The folder of examples contains sample applications demonstrating the API features of ESP-WHO.

Take one Face Detection as an example.

  1. Get into one example folder esp-who/examples/single_chip/camera_with_command_line.
cd esp-who/examples/single_chip/camera_with_command_line
  1. Compile with the latest ESP-IDF. You may run make defconfig before make flash to get the default configurations.
make defconfig
make flash
  1. Advanced users can modify some options by using the command make menuconfig.

Check README.md of each example for more details.

Resources