Odin

Pose estimation-based tracking and counting of people in videos

Demo

Demo on YouTube
Paper Abstract

Usage

Uses Docker

Pull docker image

$ docker pull jgravity/tensorflow-opencv:odin
$ docker run -it --name odin jgravity/tensorflow-opencv:odin bin/bash

Download/Install code

# git clone https://github.com/PJunhyuk/people-counting-pose
# cd people-counting-pose
# chmod u+x ./compile.sh && ./compile.sh && cd models/coco && chmod u+x download_models_wget.sh && ./download_models_wget.sh && cd -

Download sample videos in testset

# cd testset && chmod u+x ./download_testset_wget.sh && ./download_testset_wget.sh && cd -

Tracking people

# python video_tracking.py -f '{video_file_name}'

Qualified supporting video type: mov, mp4
You have to put target video file in ./testset folder

Arguments

-f, --videoFile = Path to Video File
-w, --videoWidth = Width of Output Video
-o, --videoType = Extension of Output Video

Example

# python video_tracking.py -f 'test_video_01f.mov'

Check results

> docker cp odin:/people-counting-pose/testset/{video_file_name} ./

Just get pose of people (without tracking)

# python video_pose.py -f '{video_file_name}'

Qualified supporting video type: mov, mp4

Dependencies

Use Docker jgravity/tensorflow-opencv,

or install

python 3.5.3
opencv 3.1.0
jupyter 4.2.1
git 2.1.4
tensorflow 1.3.0
pip packages

scipy 0.19.1
scikit-image 0.13.1
matplotlib 2.0.2
pyYAML 3.12
easydict 1.7
Cython 0.27.1
munkres 1.0.12
moviepy 0.2.3.2
dlib 19.7.0
imageio 2.1.2

Results (time required)

Check results_log

Reference

Test dataset

testset/test_video_01: Pedestrian overpass - original video (sample) - BriefCam Syndex
testset/test_video_02: Pedestrian Walking and Traffic Exit,Human Activity Recognition Video ,DataSet By UET Peshawar

Citation

@inproceedings{insafutdinov2017cvpr,
    title = {ArtTrack: Articulated Multi-person Tracking in the Wild},
    booktitle = {CVPR'17},
    url = {http://arxiv.org/abs/1612.01465},
    author = {Eldar Insafutdinov and Mykhaylo Andriluka and Leonid Pishchulin and Siyu Tang and Evgeny Levinkov and Bjoern Andres and Bernt Schiele}
}

@article{insafutdinov2016eccv,
    title = {DeeperCut: A Deeper, Stronger, and Faster Multi-Person Pose Estimation Model},
    booktitle = {ECCV'16},
    url = {http://arxiv.org/abs/1605.03170},
    author = {Eldar Insafutdinov and Leonid Pishchulin and Bjoern Andres and Mykhaylo Andriluka and Bernt Schiele}
}