ESP32-CAM Edge Impulse Object Detection

This project uses an ESP32-CAM (AI Thinker) module to run an Edge Impulse–trained AI object detection model and stream the results live through a web browser.

Camera frames are processed directly on the ESP32, objects are detected using the AI model, and bounding boxes with labels and confidence scores are drawn on the live video stream.

Contents

• Features
• Hardware Requirements
• Known Limitations
• Installation
• License
• Need help

Features

• ESP32-CAM (AI Thinker) support
• Object Detection with Edge Impulse
• Bounding boxes drawn on live video
• Built-in web server
• Real-time MJPEG streaming (/stream endpoint)

Hardware Requirements

• ESP32-CAM (AI Thinker)
• USB-to-TTL adapter (FTDI, CP2102, etc.)
• Jumper wires

Known Limitations

• Higher resolutions reduce FPS
• Many detections increase RAM usage
• ESP32-CAM has limited compute resources for large models

Installation

Getting Ready

1. Open Arduino IDE -> File -> Preferences
2. Add following URL to Additional boards manager URLs;

https://raw.githubusercontent.com/espressif/arduino-esp32/gh-pages/package_esp32_index.json

Collecting Data to Train the AI Model

1. Complete the Getting Ready steps
2. Open Arduino IDE -> Library Manager
3. Install the EloquentEsp32Cam library
4. Go File -> Examples -> EloquentEsp32Cam -> Collect_Images_for_EdgeImpulse
5. Upload the code to tje ESP32-CAM (For programming, see Programming ESP32-CAM)
6. Collect images of the object to be detected (collect at least 50 images)

Programming ESP32-CAM

The ESP32-CAM does not include an integrated USB-to-serial programmer so an external FTDI (USB-to-TTL) programmer is used for uploading code. (You can also use an Arduino Uno as a USB-to-serial adapter. For more information, refer to tutorial videos on YouTube.)

Train an AI model with Edge Impulse

1. Go to https://studio.edgeimpulse.com/
2. Create new project
3. Navigate to Data acquisition -> Add data -> Upload data
4. Select the folder containing the collected images
5. Go to label section and choose Enter label
6. Enter your object name
7. Click on upload data button
8. Go Labelling Queue and label your images
9. Navigate to Create Impulse
10. Add an Image Processing block
11. Add an Object Detection learning block
12. Save the impulse
13. Go to Image and set Color depth to Grayscale
14. Save parameters and click Generate Features
15. Go to Object Detection and set the Learning rate to 0.01
16. Click Start Training
17. After training is complete, go to Deployment
18. Select Arduino Library as the deployment option
19. Set the target to Espressif ESP-EYE (ESP32 240MHz)
20. Click Build and download the generated library
21. In Arduino IDE, go to Sketch → Include Library → Add .ZIP Library
22. Open ObjectDetection.ino and replace

#include <Maden_suyu_detection_inferencing.h>

with your own model header file

Update the WiFi credentials in the code before uploading it to the ESP32-CAM:

const char *ssid = "*******";
const char *password = "*******";

23. Upload the code and open the Serial Monitor at 115200 baud and copy the ip address
24. Open a browser and navigate to:

http://ESP32_IP_ADDRESS/

You will see the live camera stream with detected objects highlighted.

License

This project is licensed under the MIT License - see the LICENSE file for details.

Need Help

If you need any help contact me on LinkedIn.

⭐ If you like this project, don’t forget to give it a star on GitHub!