Système de détection d'objets en temps réel avec ESP32-CAM et YOLOv3

📷 Real-Time Object Detection System with ESP32-CAM and YOLOv3

A real-time object detection system using ESP32-CAM camera and YOLOv3 model with OpenCV.

🎯 Features

• ✅ Real-time object detection via ESP32-CAM
• ✅ YOLOv3 pre-trained model implementation
• ✅ Detection of 80 object classes (people, animals, vehicles, etc.)
• ✅ Bounding boxes and confidence scores display
• ✅ Live visualization interface with OpenCV
• ✅ Smart text positioning (always visible even for partially off-frame objects)
• ✅ Adjustable window size for comfortable viewing

📋 Prerequisites

Hardware

• ESP32-CAM AI-Thinker (with OV2640 module)
• FTDI programmer or USB-TTL cable
• WiFi network
• Computer with Python 3.7+

Software

• Arduino IDE or PlatformIO
• Python 3.7 or higher
• Required Python libraries (see Installation)

🚀 Installation

1. ESP32-CAM Setup

a) Install ESP32 Board in Arduino IDE

1. Open Arduino IDE
2. Go to File > Preferences
3. Add this URL to "Additional Board Manager URLs":

   https://dl.espressif.com/dl/package_esp32_index.json
   

4. Go to Tools > Board > Boards Manager
5. Search for "ESP32" and install ESP32 by Espressif Systems

b) Upload Code to ESP32-CAM

1. Open the esp32cam_server.ino file
2. Modify WiFi credentials:

   const char* WIFI_SSID = "YOUR_WIFI_NAME";
   const char* WIFI_PASS = "YOUR_PASSWORD";

3. Select board: Tools > Board > ESP32 Arduino > AI Thinker ESP32-CAM
4. Connect FTDI programmer:
   • FTDI 5V → ESP32 5V
   • FTDI GND → ESP32 GND
   • FTDI TX → ESP32 U0R
   • FTDI RX → ESP32 U0T
   • ESP32 IO0 → ESP32 GND (for programming mode)
5. Click Upload
6. After upload, disconnect IO0 from GND and press RESET

c) Get ESP32-CAM IP Address

1. Open Serial Monitor (115200 baud)
2. Press RESET button on ESP32-CAM
3. Note the IP address displayed (e.g., 192.168.1.16)

2. Python Environment Setup

a) Install Python Dependencies

pip install numpy opencv-python urllib3

b) Download YOLOv3 Files

Create a project folder and download these files:

1. COCO Classes File (coco.names.txt):

wget https://raw.githubusercontent.com/pjreddie/darknet/master/data/coco.names -O coco.names.txt

2. YOLOv3 Configuration (yolov3.cfg):

wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3.cfg

3. YOLOv3 Weights (~236 MB):

wget https://pjreddie.com/media/files/yolov3.weights

Or manually download from:

• YOLOv3 Weights
• YOLOv3 Config
• COCO Names

3. Project Structure

Your project folder should look like this:

esp32cam_yolo_detection/
├── esp32cam_server.ino          # ESP32-CAM Arduino code
├── object_detection_yolo.py     # Python detection script
├── yolov3.cfg                   # YOLOv3 configuration
├── yolov3.weights               # YOLOv3 pre-trained weights
└── coco.names.txt               # 80 object classes names

🎮 Usage

1. Start ESP32-CAM

• Power on ESP32-CAM
• Wait for WiFi connection (check Serial Monitor)
• Note the IP address

2. Update Python Script

Edit object_detection_yolo.py and change the URL:

url = 'http://YOUR_ESP32_IP/cam.jpg'  # Replace with your ESP32-CAM IP

3. Run Detection

python object_detection_yolo.py

4. Controls

• Window will display live feed with detected objects
• Press 'Q' to quit the application
• Resize window manually if too large

⚙️ Configuration

Adjust Detection Parameters

In object_detection_yolo.py:

# Detection confidence threshold (0.0 to 1.0)
confThreshold = 0.5  # Lower = more detections, more false positives

# Non-maximum suppression threshold
nmsThreshold = 0.3   # Lower = fewer overlapping boxes

# Window size adjustment
im_resized = cv2.resize(im, (0, 0), fx=0.5, fy=0.5)  # 0.5 = 50% size

Change Frame Size (ESP32-CAM)

In esp32cam_server.ino:

config.frame_size = FRAMESIZE_SVGA;  // Options:
// FRAMESIZE_QQVGA  (160x120)
// FRAMESIZE_QVGA   (320x240)
// FRAMESIZE_VGA    (640x480)
// FRAMESIZE_SVGA   (800x600)
// FRAMESIZE_XGA    (1024x768)
// FRAMESIZE_UXGA   (1600x1200)

🐛 Troubleshooting

ESP32-CAM Issues

Problem: Camera initialization fails

Solution: Check all pin connections, ensure OV2640 camera is properly seated

Problem: Can't connect to WiFi

Solution: 
- Verify WiFi credentials
- Check if WiFi is 2.4GHz (ESP32 doesn't support 5GHz)
- Check router firewall settings

Problem: Image quality is poor

Solution: Adjust JPEG quality in code:
config.jpeg_quality = 10;  // Lower = better quality (10-63)

Python Script Issues

Problem: "Connection refused" or timeout

Solution:
- Verify ESP32-CAM IP address
- Ping ESP32-CAM: ping 192.168.1.16
- Check firewall settings
- Ensure both devices are on same network

Problem: OpenCV window doesn't appear

Solution:
- On Linux: sudo apt-get install python3-opencv
- On macOS: brew install opencv
- Try: cv2.namedWindow('ESP32-CAM Detection', cv2.WINDOW_NORMAL)

Problem: "Failed to load YOLO files"

Solution:
- Verify all three files exist (cfg, weights, names)
- Check file names exactly match
- Re-download yolov3.weights if corrupted

Problem: Slow detection/Low FPS

Solution:
- Reduce frame size on ESP32-CAM
- Reduce input size: whT = 320 (or lower to 256)
- Use YOLOv3-tiny for faster detection (smaller model)

🔧 Advanced Features

Use YOLOv3-Tiny (Faster, Less Accurate)

For faster detection on slower computers:

1. Download YOLOv3-Tiny files:

wget https://pjreddie.com/media/files/yolov3-tiny.weights
wget https://raw.githubusercontent.com/pjreddie/darknet/master/cfg/yolov3-tiny.cfg

2. Update Python script:

modelConfig = 'yolov3-tiny.cfg'
modelWeights = 'yolov3-tiny.weights'

Detect Specific Objects Only

Modify the findObject function to filter specific classes:

# Only detect people and cats
allowed_classes = ['person', 'cat']

if classNames[classIds[i]] in allowed_classes:
    # Draw box and label
    cv2.rectangle(im, (x, y), (x + w, y + h), (255, 0, 255), 2)
    # ... rest of the code

Save Detections to Video

Add this code to save output:

# Before the while loop
fourcc = cv2.VideoWriter_fourcc(*'XVID')
out = cv2.VideoWriter('output.avi', fourcc, 20.0, (640, 480))

# Inside the loop, after findObject()
out.write(im_resized)

# After the loop
out.release()

📊 Detected Object Classes

The system can detect 80 classes from COCO dataset:

• People: person
• Animals: cat, dog, bird, horse, sheep, cow, elephant, bear, zebra, giraffe
• Vehicles: car, motorcycle, airplane, bus, train, truck, boat
• Indoor: chair, couch, bed, dining table, toilet, tv, laptop, mouse, keyboard
• Food: banana, apple, sandwich, orange, broccoli, carrot, hot dog, pizza, donut, cake
• And many more...

📝 Code Explanation

ESP32-CAM Code

handleJpg()  // Captures and streams JPEG image via HTTP

Python Detection Script

cv2.dnn.readNetFromDarknet()  // Loads YOLOv3 model
cv2.dnn.blobFromImage()       // Preprocesses image for YOLO
net.forward()                  // Runs object detection
cv2.dnn.NMSBoxes()            // Removes duplicate detections

🤝 Contributing

Contributions are welcome! Please feel free to submit a Pull Request.

📄 License

This project is open source and available under the MIT License.

🙏 Acknowledgments

• YOLOv3: Joseph Redmon and Ali Farhadi
• ESP32-CAM: Espressif Systems
• OpenCV: Open Source Computer Vision Library
• COCO Dataset: Common Objects in Context

📧 Contact

For questions or issues, please open an issue on GitHub.

🔗 Useful Links

• YOLOv3 Paper
• ESP32-CAM Documentation
• OpenCV Documentation
• COCO Dataset

---

⭐ If you find this project helpful, please give it a star!