tomdevelops / YOLOv5s-pytorch-deepsort-passenger-counting

YOLOv5 + Deep Sort Passanger Counter by Heads

A computer vision and deep learning project for counting and tracking passengers using head recognition.

This project was developed by both Swiss and Chinese students as part of a joint venture between University of Applied Science Northwestern Switzerland FHNW and Shenzhen Technology University SZTU.

Standard	Custom

The main task includes to count passenger flows in a metro station. Therefore YOLOv5 and Deep Sort as well as a counting mechanism were merged (from already existing Repos). In addition, new weights of both YOLO and the feature extractor of Deep Sort were trained using custom datasets.

The following repos were adapted and reused for this purpose:

YOLOv5 + Deep Sort Baseline Implementation

Setup + Training of a Siamese Network for Custom Head Tracking

YOLOv5

Deep Sort

Deep Sort Custom Head Dataset

Dataset Prep

Installation

First, clone the repository:

git clone [e-mail]:bill-lou/YOLOv5s-pytorch-deepsort-passenger-counting.git

Use this command to install all the necessary packages. Note that we are using python3

pip install -r requirements.txt

Configuration

There are two wheights available for YOLOv5, make sure to adjust the parser argument in count.py befor running the code

You can Download the Weights from:

Standard YOLO

Custom Head

yolov5x.pt for standard YOLO Detection

parser.add_argument('--weights',
                    type=str,
                    default='yolov5/weights/yolov5x.pt',
                    help='model.pt path')

custom_train_best.pt for Custom Head Detection

parser.add_argument('--weights',
                    type=str,
                    default='yolov5/weights/custom_train_best.pt',
                    help='model.pt path')

Download as well the Deep Sort Checkpoints and add it to the folder deep_sort_pytorch\deep_sort\deep\checkpoint

You can Download the Checkpoints from:

Standard Deep Sort

Custom Head Deep Sort

IN and Out Folder

Input Video Stream Add your input Stream (Video File) into the in folder

Output Video Stream Once the code is successfully completed - you should find a video of your input + stream + the augmentend information of tracker, detector and counting in the out folder

Run the Code with a Counter

Counting Line + YOLO + Deep Sort Standard

python3 count.py --deepsort 1

Counting Line + YOLO + Deep Sort Custom

python3 count.py --deepsort 2

Run the Code without a Counter

YOLO + Deep Sort Standard

python3 track.py --deepsort 1

YOLO + Deep Sort Custom

python3 track.py --deepsort 2