We present a tool to synchronize videos of people performing similar actions. This can be useful for applications like -
- Comparing game play of two players
- Side by side view of dance moves
- Comparison of any moves that require side by side video
This open source tool allows you to compare any two videos by bringing them in sync using the state of the art models at its backend for performing pose estimation and matching the poses.
It consists of three stages:
- Cropping (Using YOLO v5 / tracker)
- Pose Detection (Movenet)
- Video Matching (Dynamic Time Warping)
-
Cropping
First, video is cropped frame by frame with help of object detection which is supported by two methods, YOLO and Tracker by OpenCV.
-
Pose Detection
Then, MoveNet Pose Detection is applied on cropped video which returns the keypoints.
-
Video Matching
Finally, DTW (Dynamic Time Warping) processes the sequences of pose detected from two videos.
We published the paper explaining our approach in detail, on arxiv.org which can be found at https://arxiv.org/pdf/2308.12600.pdf
Check out the demo of PoseSync on HuggingFace and share the amazing results with us : https://huggingface.co/spaces/infocusp/PoseSync-Video-Matching-Tool
git clone https://github.com/InFoCusp/posesync.git
pip install -r requirements.txt
python3 main.py
@article{javia2023posesync,
title={PoseSync: Robust pose based video synchronization},
author={Javia, Rishit and Shah, Falak and Dave, Shivam},
journal={arXiv preprint arXiv:2308.12600},
year={2023}
}