Turning_Movement_Counter_w/_Yolov7_StrongSORT_OSNet

----------------------- Web Application ------------------------------------------- Output Video --------------------

Introduction

This is a work in progress

This repository is a project that aims to count the number of vehicles making turning movements at an intersection. It leverages the powerful capabilities of the Yolov7_StrongSORT_OSNet framework to achieve this goal. The project includes a user-friendly web application that allows users to define the different movements at the intersection. A simple algorithm is employed to count the vehicles in 15-minute increments, and object tracking is achieved through the use of Yolov7_StrongSORT_OSNet. A Turning Movement Count, is commonly used in traffic engineering to model intersections and optimize signal timing.

Prerequisites

In order to run the colab example you need to first have an ngrok auth token. No need for ngrok if you want to run on your local machine.

Before you run the tracker

1. Clone the repository:

git clone https://github.com/joshkuminski/Turning_Movement_Counter_Yolov7_StrongSort_OSNet.git

2. cd into repo and clone the multiple object tracking (MOT) project:

cd Turning_Movement_Counter_with_Yolov7_StrongSORT_OSNet
git clone https://github.com/joshkuminski/Yolov7_StrongSORT_OSNet.git

3. cd into the MOT repo and clone the yolov7 project:

cd Yolov7_StrongSORT_OSNet
git clone https://github.com/joshkuminski/yolov7.git

4. clone the ReID project:

cd strong_sort\deep\reid
git clone https://github.com/joshkuminski/deep-person-reid.git

5. Make sure that you fulfill all the requirements: Python 3.8 or later with all requirements.txt dependencies installed, including torch>=1.7. If using a gpu for tracking, you need to uncomment the torch install lines in requirements.txt if gpu is not available. Cuda toolkit is required to utilize gpu. To install, run:

cd Turning_Movement_Counter_with_Yolov7_StrongSORT_OSNet/local_requirements
pip install -r requirements.txt
pip install -r requirements_gpu.txt

Run the Flask App

cd Flask_App
$ python main.py

Run the Tracker

$ python track.py --source <path to video> --yolo-weights yolov7-e6e.pt --img 640 --classes 2 3 5 7 --strong-sort-weights osnet_x0_25_market1501.pt --save-vid
                                                                                                                                                     --show-vid --device 0 #if cuda is available

Custom Dataset

Custom dataset created for vehicle detection only. This dataset is more accurate for turning movement counts. The custom classes available are [car, truck, school bus, person, trailer, bicycle]. Contact me for the custom yolov7 weight file or if you would like to contribute to the dataset.

Contact

For questions please email [email protected] For bugs and feature requests please visit GitHub Issues.