Testing but not recognizing any body

[Bardo91]

Hi, I am testing your code but I am not able to make it work with me. I did some modifications to make it work with catkin and ros-kinetic in ubuntu 16.04 and OpenCV-3.3.1, but haven't touch the code. This is what I got:

I also tried it with the dataset you provide and works randomly:

Could you give me any advice?

[mgb45]

Hi,

This code isn't really maintained at this point, since there are many more effective approaches to human pose estimation available nowadays, for example this. The primary contribution of this work was the 2D to 3D pose tracking node

_mkfbodytrackerpdaf

This node takes a subset of 2D joint measurements and users a pose tracking approach to infer the 3D pose of the person. The primary limiting factor for this node is the input 2D measurements provided by the face and hand detection - here I just use some carefully tuned ones for demonstration. Judging by your picture, this is what seems to be failing - the face detector isn't picking you up, which is why the pose is failing. You could try modifying the parameters in the face detector node,

facetracking

but ideally you want to replace the measurement nodes with something more modern, like a convnet joint detector that you can sample from to feed into the tracker.

If you can share a bag file I can probably tune the parameters to make these work for you, but as I mentioned the input measurement nodes were never intended to provide a stable and robust solution, only a demonstration of the pose tracker.

[Bardo91]

Hi, Thank you very much for your answer. I know there are other methods using DNN, I was inspecting other possibilities that I found :). I will try to feed your module as you said with the result of another detector. Thanks again. I am closing the issue