Object Detection in Video

Overview

This project implements an object detection system using the DETR (DEtection TRansformer) model from Hugging Face's Transformers library. It processes video frames to detect specific restricted classes of objects and draws bounding boxes around them.

Features

Detects specific restricted objects: person, cell phone, laptop, TV, keyboard, and mouse.
Draws bounding boxes around detected objects with different colors.
Saves processed frames with detections to a temporary directory.

Requirements

To run this project, you'll need the following Python packages:

torch
transformers
Pillow
opencv-python
numpy

Setup Instructions

Installing FFmpeg

If you are using macOS, you can install FFmpeg using Homebrew. Open your terminal and run the following command:

brew install ffmpeg

Creating a Conda Environment

Create a new Conda environment:

Open your terminal (or Anaconda Prompt) and run the following command:
```
conda create --name video_analyzer python=3.11
```
Replace video_analyzer with your preferred environment name if needed.
Activate the Conda environment:
```
conda activate video_analyzer
```
Install required packages:

Once the environment is activated, install the required packages using the requirements.txt file provided in the project:
```
pip install -r requirements.txt
```

Usage

Running the Script

To run the object detection on a video file, execute the script from the command line:

python main.py <path_to_video_file> --frame_rate=<frame_rate> --display_video=<display_video> --store_image_path=<image_path> --store_video_path=<video_path>

Arguments

<path_to_video_file>: The path to the input video file.
--frame_rate: (Optional) The rate at which frames are extracted from the video. The default value is 1 frame per second.

Example

python main.py input_video.mp4
python main.py input_video.mp4 --frame_rate=1
python main.py input_video.mp4 --frame_rate=1 --display_video=True
python main.py input_video.mp4 --frame_rate=1 --store_image_path=/tmp/ai_files
python main.py input_video.mp4 --frame_rate=1 --store_video_path=/tmp/ai_files
python main.py input_video.mp4 --frame_rate=1 --store_image_path=/tmp/ai_files --display_video=True
python main.py ~/Movies/video_for_ai2.mp4 --frame_rate=1 --store_video_path=/tmp/ai_files  --display_video=True --store_image_path=/tmp/ai_files
python -m unittest discover -s tests

This command will process input_video.mp4, extracting 2 frames per second. The output frames with detected objects will be saved in the temporary directory /tmp/ai_files.

Output

Processed frames will be saved in the specified temporary directory with filenames in the format detected_frame_<index>_<timestamp>.png, where <index> is the frame index and <timestamp> indicates when the frame was processed.

Code Structure

detection_model.py: Contains the implementation of the DETR model and other related classes for object detection.
frame_processor.py: Handles the extraction and processing of video frames.
detection_drawer.py: Manages the drawing of bounding boxes and labels on the video frames.
main.py: The entry point of the application.

Author

Kinn Coelho Juliao kinncj@gmail.com

License

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

kinncj/object-detection