This repo documents our research in the most recent mocap approaches. These 3 approaches were the most successful:

1: OpenPose Multiperson (found in this repo in the notebook titled Test.ipynb) The code belongs to the blog https://www.learnopencv.com/multi-person-pose-estimation-in-opencv-using-openpose and was modified in Test.ipynb to accomodate web cam multiperson pose detection (instead of image-only pose detection)

A.Requirements :

1. OpenCV > 3.4.1
2. Matplotlib for Notebook
3. RUN getModels.sh from command line Or Download caffe model from http://posefs1.perception.cs.cmu.edu/Users/ZheCao/pose_iter_440000.caffemodel and put it in pose/coco folder

2: PIFuHD We changed the code from this paper https://ai.facebook.com/research/publications/pifuhd-multi-level-pixel-aligned-implicit-function-for-high-resolution-3d-human-digitization/ which constructs 3d models with fine details including occluded parts, to accomodate video input. * https://colab.research.google.com/drive/1pSU9msr7nXHpllaCssvBFa5_tRl3oWTP?usp=sharing#scrollTo=z7pHnWccixto

3: lightweight-human-pose-estimation We also tried the code from this repo https://github.com/Daniil-Osokin/lightweight-human-pose-estimation-3d-demo.pytorch and it was ran successfully.