New project for OpenCap, which is a new software package to estimate 3D human movement dynamics from smartphone videos. OpenCap strongly relies on OpenSim.
-
Mahsa Agz
- Posts: 2
- Joined: Mon Nov 13, 2023 10:00 pm
Post
by Mahsa Agz » Wed Nov 15, 2023 7:10 pm
Hello,
On November 13th, we conducted experiments involving tasks like "moving arms," "squatting," and "jumping," both with and without occlusion. The processing time for the videos was generally reasonable, averaging around 1 minute and 35 seconds in both conditions. However, in both conditions (with and without occlusion) we encountered a couple of issues with the skeleton. In almost all tasks (simple and complex) the body was shaky, especially in the upper body. And arms had spurious flexion and the body was twisted. Additionally, in videos with occlusion, the body seemed to lean forward.
We would like to know what are the possible reasons for these problems and how we can have a better performance.
These are examples:
-
Scott Uhlrich
- Posts: 124
- Joined: Tue Jan 20, 2015 4:29 pm
Post
by Scott Uhlrich » Wed Nov 15, 2023 10:20 pm
For upper body kinematics, Id recommend using the hrnet pose model and the musculoskeletal model with ISB shoulder. These can be selected in Advanced Settings on the neutral pose page. This may solve the first two issues.
Generally when the kinematics are inaccurate, it can be helpful to download the data and visualize the marker data in the OpenSim gui to see which keypoint position is causing the problem.
OpenCap needs at least 3 cameras to see every body part, so occlusions like shown will not work.
