When setting up a markerless motion capture system, camera placement can significantly affect the accuracy of results. This is no different from marker-based modalities, and, like marker-based systems, a fair bit of time and effort is required in order to properly set up the cameras. The purpose of this post is to examine the differences between the system setups, and try to provide some information on how to best set up the system.
How do we set up the cameras? In a lot of cases, these systems are being added to an existing marker-based setup, so the first thought is to place the cameras directly beside the marker-based cameras. Although this is very convenient, because the cameras are typically mounted on a truss-like system around the lab, with existing conduits for wires, it does not represent an even close to ideal setup for video cameras.
The first difference is the size of the sensor between the systems. Often the marker-based system has a 10X larger sensor size than the video camera, so they can capture more granular data. The second difference is the requirement for focus and detail in the image. A colleague of mine said to me recently “we aren’t just picking up white dots in an image”, which pretty much sums it up. The lenses on these video cameras are rarely zoom lenses, so it’s highly unlikely that they will meet the requirement for focus when they are positioned close to a completely different camera. Because the features we are looking for are rich (not white dots), the quality of the input image can greatly affect the results.
To further this, with the help of a collaborator, we investigated the effect of camera placement on gait kinematics. This will eventually become a full paper, so here I’m just presenting a small subset of what we found. We tested two different environments, one with the cameras high on a truss system, and one more in line with our requirements (cameras surrounding the participant at mid height). The data was compelling.
The implications of this are that researchers really need to pay attention to their setup. Just because there is existing infrastructure does not mean that the setup is positioned ideally, and the same, if not more care is required to get quality images. However, like anything, once the team understands the requirements, it is easy to set up the system. The same collaborators who collected these data can set up their system in around 10 minutes. According to them, the biggest challenge was that they had to unlearn everything they used to apply to marker-based system setups, because they just follow different rules.
As a final aside, we are very happy to provide recommendations on how to set up the camera systems (even if we don’t sell cameras) for your environment, just beware that we will likely recommend not using the truss!