How to collect good data while performing markerless and marker-based mocap.
Like any measurement system, there are many factors that can influence the quality of the output data from Theia3D. For example, dark lighting within the space can make feature detection difficult (if we can’t see it, how can we measure it?), a poor calibration can make 3D consolation inaccurate, or sub-optimal camera placement can result in redundant triangulation from similar views.
A lot of these difficulties are also common to marker based motion capture systems. For example, the operator needs to avoid glare that might interfere with marker detection, the calibration needs to be spot on, and markers must be placed in precisely the same location for each session. In this collection modality, any of these issues can result in marker dropout which may jeopardize the entire collection.
In machine learning driven modalities, the dropout problem is a bit less obvious. Though algorithms are used to parse out good and bad data via statistical analysis, there is considerable overlap between a bad detection that makes sense, versus a bad detection that is caused by setup error. Like any mocap system, this requires patience, attention to detail and a bit of experience with the system.
The University of Central Lancashire's setup demonstrating good lighting and camera positioning for Theia3D.
Below we have outlined some key recommendations that will help ensure your setup and first few data collections run smoothly:
Pay attention to the details during setup: We have recommendations on where cameras should be placed, so please take time to understand these recommendations and follow them. For example, using a bird’s eye view or an existing truss system isn’t ideal for Theia3D. Just because there is an existing infrastructure for camera mounting does not mean it should be used. Check out our additional recommendations here.
Calibration, calibration, calibration: Make sure your calibration residuals are small and that you are using the checkerboard and not the object to calibrate. If you need to use the object, make sure the object measurements and their locations can be selected precisely (~ 1mm). Using a wand and L-frame is also a suitable calibration method if you are using a Vicon or Qualisys system. We have detailed information on how to calibrate using the checkerboard on our Youtube channel here. Don’t forget to make sure the board is visible when it is set on the ground. It’s the intersection of the two blue squares that are used as the global coordinate system.
Clothing and apparel: Make sure you pay attention to clothing and apparel. In the same way that very dark and very light lighting can make features less rich, the same considerations should be applied to clothing. We have a previous blog post on how clothing can impact data processing in Theia3D, and the HMRL from Queen's University has also published on this topic.
Know your software: There are a ton of features within Theia3D that provide flexibility regarding model scaling, filtering, and joint constraints; we strongly recommend that you read these! Too often we hear that the user does not have control over how the model is scaled, and this is simply not true.
Developing expertise using a markerless system takes some time. Please be patient and know that you can always reach out to our support team if you have any questions (support@theiamarkerless.ca). Our hope is that these challenges can be avoided by other experimenters, and that their data collections can go smoothly (unlike ours in the early days). While it is possible to collect data very quickly, we recommend that you go slow in the beginning to make sure you avoid some of the common pitfalls described above.
To learn more about Theia3D, click here to book a demo.