Post Estimation using Open Pose

8th Jul 2020

Having spent some time with my Vive tracker based motion capture setup I'm again thinking "What would it look like if it was simple?" Doing a complete motion capture session is hardware intesive, and time consuming, What if I could just video an actor and convert that to a motion capture skeleton later? That is the vision of the Open Pose software. The Github for Openpose makes getting the software running on Windows extremely easy.

Openpose breaks the body down into face, hands and general skeleton. I'm mostly interested in the skeleton tracking. Unfortunately Openpose doesn't seem to give a way to convert the skeleton tracking to a standard FBX animation file or even BVH. The output is per frame JSON, so a conversion is necessary. This person on Youtube is doing some work with the face tracking features, and he has software to convert the JSONs to Blender, but only for the face.

It looks like MoCapNET Project has a OpenPose Json to BVH converter that might work. No support for windows though, I'll jump over to my Linux box to see if I can get that working.

Setting Up MocapNet

Grab the init script for Ubuntu here

chmod +x and run

CentOS Example

Here are the steps I took to get this running on Centos 7. Using the Docker image was pretty straight forward.

You need to setup Docker and NVIDIA container toolkit.

Startup the Docker:

sudo docker run -it -v /home/rw/Pictures/:/openpose/data/ --net=host -e DISPLAY --runtime=nvidia exsidius/openpose

Run openpose on a video input file:

build/examples/openpose/openpose.bin --video /openpose/data/bela_strut.mov

There are some good examples from CMU-Perceptual Computing Lab I am mostly interested in getting the body skeletal tracking, some examples focus on the facial capture.

There is very little info on converting these JSON files to FBX or BVH. MocapNET seems to be the only one I've found

Note on compiling MocapNet on Centos7: After installing cmake3 you need to edit MocapNet's CMakelists.txt and add gnu99 compiler flag "set(CMAKE_C_FLAGS "-fPIC -s -std=gnu99") "