Skip to content

Latest commit

 

History

History
87 lines (59 loc) · 4.11 KB

README.md

File metadata and controls

87 lines (59 loc) · 4.11 KB
Raw

Resource-Constrained Intelligent Trap- An Insects monitorning system using ESP32 and SSD-MobileNetV2 FOMO model

RCIT, boasting advanced functionalities while conserving minimal resources on the ESP-EYE, not only achieves an impressive 92% F1 Score on the validation dataset but also operates efficiently with a mean processing time of just 6 seconds per image and peak RAM usage of 2.4Mb per task. Furthermore, the project encompasses a user-friendly dashboard built on Firebase as its database and utilizes Node-RED as the primary monitoring server.

Requirements

Hardware

  • Espressif ESP32 based development boards, preferably ESP-EYE (ESP32) and FireBeetle Board (ESP32). Using with other boards is possible, but code modifications is needed. For more on that read Using with other ESP32 boards.

Tools

  • Install ESP IDF v4.4, following the instructions for your OS from this page.
  • Clone this repository of course!?

Build your own model?

If you are using Edge Impulse, do the following steps:

  • Go to Deployment, select and build model as C++ library, a folder will be downloaded automatically to your computer.
  • Open the downloaded folder, copy three folders edge-impulse-sdk, model-parameters and tflite-model and paste to this repository.

Building the application

  • First, open the file \edge-impulse\inference\ei_run_impulse.h and configure your Firebase project information:
// create a new api key and add it here  
#define API_KEY "AIzaSyCwBgvqH-j_TFMXijSNxnI2lC4f_ABCDEF"

// Copy your firebase real time database link here 
#define DATABASE_URL "https://smarttrap2022-XXXXX-default-rtdb.firebaseio.com/"  

#define USER_EMAIL "quannm293@gmail.com"  // This gmail does not exist outside your database. it only exists in the firebase project as a user
#define USER_PASSWORD "123456788"      // Dont add your gmail credentials. Setup users authentication in your Firebase project first
  • Then from the firmware folder execute:
get_idf
clear && idf.py --no-ccache build

get_idf is an alias for export.sh script that sets up ESP IDF environment variables. Read more about it here.

  • To connect Firebase to Node-Red and create User-Dashboard. Follow this tutorial.

Flash

Connect the ESP32 board to your computer.

Run:

idf.py --no-ccache -p COM3 -b 115200 flash monitor
  • COM3 needs to be changed to actual port where ESP32 is connected on your system.
  • The partition table needs to be adjusted to be compatible with your model and device.

Serial connection

Use screen, minicom or Serial monitor in Arduino IDE to set up a serial connection over USB. The following UART settings are used: 115200 baud, 8N1.

Some experiments

These are some experiments of SSD-MobileNetV2 FOMO model deployment on ESP-EYE

Stats comparison between two SSD-MobileNetV2 FOMO model Version

Models Device Ram (Mb) Flash (Kb) Time (ms)
FOMO 0.1 ESP32 2.4 52.4 5694
FOMO 0.35 ESP32 2.4 53.3 6427

Performance comparison between two SSD-MobileNetV2 FOMO model Version

Threshold Accuracy FOMO 0.1 Accuracy FOMO 0.35
0.25 0.85 0.92
0.5 0.96 1.00
0.75 0.75 0.78

Dashboard Interface on Node-Red

Contacts

Email: quan.nm207984@sis.hust.edu.vn

Phone: 0961266396