ESP32-CAM Object Detection using Edge Impulse (with OLED)

This project demonstrates real-time object detection using the ESP32-CAM module and Edge Impulse. The system is trained to detect objects like onion, potato, and tomato, and displays detection results via Serial Monitor and OLED display.

---

Project Features

• Object detection using ESP32-CAM (AI Thinker)
• Custom dataset collected using ESP32-CAM
• Edge Impulse YOLO-based object detection
• Optimized for low memory devices
• OLED display support for live results

---

Hardware Requirements

• ESP32-CAM (AI Thinker)
• USB-to-TTL (FTDI) Programmer
• Breadboard & jumper wires
• Micro-USB cable
• Computer with Arduino IDE
• Wi-Fi connection

---

Hardware Connections

ESP32-CAM Programming Connections

ESP32-CAM	FTDI
5V	5V
GND	GND
U0R	TX
U0T	RX
GPIO0	GND (Only during upload)

Important

• GPIO0 must be connected to GND while uploading
• Remove GPIO0-GND after upload for normal operation

---

Software Requirements

• Arduino IDE
• ESP32 Board Package
• Edge Impulse Account (Free)
• Eloquent ESP32 CAM Library

---

Arduino IDE Setup

1. Install Arduino IDE

2. Add ESP32 Board Manager

3. Install ESP32 by Espressif Systems

4. Select:

   • Board: AI Thinker ESP32-CAM
   • Programmer: ESP32 Arduino
   • Baud Rate: 115200

---

Image Collection Using ESP32-CAM

1️ Install Library

Install Eloquent ESP32 CAM library from Arduino Library Manager.

2️ Open Image Capture Example

File → Examples → eloquentesp32cam → esp32cam_tp3

3️ Modify Code

• Add your Wi-Fi SSID & Password
• Set camera type to AI Thinker

#define CAMERA_MODEL_AI_THINKER

4️ Upload Code

• Connect GPIO0 → GND
• Power reset ESP32-CAM
• Upload sketch
• Remove GPIO0 → GND
• Power reset again

5️ Access Image Server

• Open Serial Monitor
• Copy the IP address
• Open it in a browser (same Wi-Fi network)

---

Dataset Collection

• Fix ESP32-CAM position
• Place object inside camera frame
• Rotate object slowly
• Capture ~50 images per object
• Download images as ZIP
• Repeat for all classes:

---

Edge Impulse Setup

1️ Create Project

• Log in to Edge Impulse
• Create new project (e.g., obj-detect)

---

2️ Upload Dataset

• Go to Data Acquisition

• Upload ZIP files

• Enable:

  • Automatically split
  • Manually label

---

3️ Label Images

• Go to Labeling Queue
• Draw bounding boxes
• Assign correct labels

---

Model Training

1️ Create Impulse

• Image size: 96 × 96

• Fit shortest axis

• Blocks:

  • Image Processing
  • Object Detection

---

2️ Generate Features

• Color depth: Grayscale
• Generate features
• Ensure clear feature separation

---

3️ Train Model

• Training cycles: 60
• Learning rate: 0.01
• Processor: CPU
• Start training
• Aim for 90–100% accuracy

---

Deployment

1️ Export Model

• Go to Deployment
• Select Arduino Library
• Target device: ESP32-EYE
• Download ZIP

---

2️ Add Library to Arduino

Sketch → Include Library → Add .ZIP Library

---

3️ Open Example Code

File → Examples → (Your_Model_Name) → esp32_camera

• Disable ESP-EYE
• Enable ESP32-CAM

---

4️ Upload Final Code

• GPIO0 → GND
• Power reset
• Upload
• Remove GPIO0
• Power reset again

---

Testing

1. Open Serial Monitor

2. Place object in front of camera

3. Observe:

   • Detected label
   • Confidence score

---

Tips for Better Accuracy

• Use consistent lighting
• Avoid cluttered backgrounds
• Collect images from multiple angles
• Keep object centered during training

---

Future Improvements

• Add speaker module for voice output
• Store MP3 files on SD card
• Autonomous robot integration
• Multi-object detection

---

Credits

• ESP32-CAM Community
• Edge Impulse Team
• Eloquent Arduino Library