scs/afz

PoC AFZ

Proof of Concept to count people going from one area to another.

Getting Started

Create

conda env create -f environment.yml
source activate afz

Labeling

python main_labeler.py

Pipeline

The pipeline consists of three main steps.

Object Detection

Using https://pjreddie.com/darknet/yolo/

Object Tracking

Using https://github.com/nwojke/deep_sort

We used the implementation of Yolov3 fr https://github.com/Qidian213/deep_sort_yolov3

Counting

Own implementation using a comparison between the area inside and outside.

DataSet

The default training of Yolo is based on the Coco Dataset. We also tried the Crowd Human Dataset.