rdesarz / cnnmot

A multiple object tracker based on CNN

CNN-MOT

The goal of this project is to implement a multiple object tracker algorithm for computer vision applications.

Object detection

The object detection algorithm is based on a CNN pre-trained Yolo v3 model proposed in this paper. Some part of the code are used in this project as well as part of this tutorial

Installation

The project is written in Python (v3.6). You can install it using pip.

pip3 install git+https://github.com/rdesarz/cnnmot.git

Keras model generation

This project is using Keras however the and so the model needs to be translated to a Keras model before using any of the scripts. Run the following instructions in the root folder of the repo. It will download the model and translate it.

wget  https://pjreddie.com/media/files/yolov3.weights
./bin/generate_keras_model.py --weights yolov3.weights --model_name yolov3.h5

Usage

Jupyter notebook

To get a demo of the detection, a Jupyter notebook yolov3_object_detection_colab_example located in the notebook folder of the project is available.

Object detection

Using webcam

./bin/webcam_object_detection.py --model yolov3.h5

Using image file

./bin/image_file_object_detection.py --model yolov3.h5 -- image example.jpg

License

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