Dynamic Capture Infer (YOLOv3 Object Detection)

A real-time object detection application using YOLOv3 implemented in C++ with OpenCV. This application can process live camera feeds or video files and detect objects in real-time.

Features

• Real-time object detection using YOLOv3
• Support for both camera input and video file input
• Multi-threaded processing for better performance
• Video recording capability
• Screenshot functionality
• Filter detection by specific object types

Requirements

• C++11 or higher
• OpenCV 4.x with DNN module
• pthread library

Installation

Install OpenCV

On Ubuntu/Debian:

sudo apt-get update
sudo apt-get install libopencv-dev python3-opencv

On macOS with Homebrew:

brew install opencv

On Windows: Download and install OpenCV from opencv.org

Compilation

Compile the program with the following command:

g++ main.cpp src/*.cpp -o output -std=c++11 `pkg-config --cflags --libs opencv` -pthread

Or if you're using a Makefile approach:

make

Usage

Basic Usage

For all labels

./output --camera 0 --label --save_image true

For only one label

./output --camera 0 --label "person" --save_image true

Camera Options

• Change camera by changing the number after --camera (e.g., --camera 1)
• During streaming, press 0/1/2 to switch cameras

Video Input

./output --video "video_name.extension" --label --save_image true

Controls

• Press ESC to stop the program
• Press a to save the current frame as an image
• Press 0, 1, or 2 to switch cameras during live streaming

Project Structure

.
├── main.cpp                 # Main application entry point
├── include/
│   ├── Queue.h              # Thread-safe queue implementation
│   ├── ObjectDetector.h     # Object detection functionality
│   ├── VideoCapture.h       # Video capture functionality
│   ├── VideoWriter.h        # Video writing functionality
│   └── Utils.h              # Utility functions
├── src/
│   ├── Queue.cpp            # Queue implementation
│   ├── ObjectDetector.cpp   # Object detection implementation
│   ├── VideoCapture.cpp     # Video capture implementation
│   ├── VideoWriter.cpp      # Video writing implementation
│   └── Utils.cpp            # Utility functions implementation
├── yolov3.cfg               # YOLOv3 configuration file
├── yolov3.weights           # YOLOv3 pre-trained weights
├── coco.names               # COCO dataset class names
└── README.md                # This file

How It Works

1. Video Capture: Captures frames from camera or video file
2. Object Detection: Processes frames using YOLOv3 neural network
3. Post-processing: Applies Non-Maximum Suppression to remove duplicate detections
4. Visualization: Draws bounding boxes and labels on detected objects
5. Output: Displays results in real-time and optionally saves video/images

Configuration

Model Files

• yolov3.cfg: YOLOv3 configuration file
• yolov3.weights: Pre-trained YOLOv3 weights (download separately)
• coco.names: Class names for COCO dataset

Download Required Files

Download the YOLOv3 weights from here and place them in the project root directory.

Troubleshooting

Compilation Issues

If you encounter compilation errors related to OpenCV headers:

1. Make sure OpenCV is properly installed
2. Try using pkg-config to get the correct flags:

   g++ main.cpp src/*.cpp -o output -std=c++11 `pkg-config --cflags --libs opencv4` -pthread

Runtime Issues

• Ensure the coco.names file is in the working directory
• Verify that the model files (yolov3.cfg and yolov3.weights) are present
• Check that camera devices are accessible

License

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

Acknowledgments

• YOLOv3 implementation based on the original Darknet model
• OpenCV for computer vision capabilities
• COCO dataset for training data