Object Tracking and Image Classification

This project utilizes OpenCV to perform object tracking and image classification in real-time using a pre-trained DenseNet-121 model and the CSRT tracking algorithm.

Features

• Real-time object tracking with CSRT (Channel and Spatial Reliability Tracker)
• Image classification using DenseNet-121, a convolutional neural network model
• Automatic extraction and classification of objects within a user-defined region of interest

Prerequisites

Before you run this project, you must have the following installed:

• Python 3.6 or higher
• opencv-python
• opencv-contrib-python
• NumPy

Installation

Clone this repository to your local machine using:

git clone https://github.com/chenyu-chang/object_traking_opencv

Ensure you have the required packages:

pip install numpy opencv-python opencv-contrib-python

Usage

Run the script using Python:

python object_tracker.py

Upon running the script, a video window will open. Use the mouse to select a region of interest (ROI) for tracking and classification. The tracker will then initialize and begin tracking the object, with its predicted class displayed on the screen.

Model Files

The DenseNet-121 model requires two files:

• DenseNet_121.prototxt: The architecture of the CNN.
• DenseNet_121.caffemodel: The pre-trained weights.

Make sure to download these files and place them in the densenet_121 directory before running the script. The synset_words.txt file containing class labels should also be in this directory. You can download by link: download

Additional Notes

• The ROI selection window can be closed by pressing the 'Esc' key.
• Tracking and classification continue in real-time; to exit, press the 'Esc' key in the main tracking window.

Acknowledgements

• DenseNet-121 model creators and maintainers.
• OpenCV contributors and maintainers.