3. Data Collection¶
3.1. Data Collection Principles¶
The following guidelines describe data collection best practices, but are not exhaustive. You may obtain high quality markerless motion capture data under different conditions from those described below.
Camera Setup
Camera system setups differ from location to location, and may be subject to challenging data collection environment constraints. General recommendations for setting up your camera system include:
Cameras as close as possible to the capture volume, while ensuring the entire capture volume remains within view for all cameras.
Cameras 4-8 feet (1-2.5 m) above the ground.
Avoid partial views of subjects, such as lower body or upper body only.
Avoid unusual camera views, especially those positioned very high or very low, or extremely tilted.
Aim for a symmetrical camera setup that surrounds the entire capture volume, such as a circle, oval, or rectangle.
Camera Settings
The camera settings you choose will depend on the movements being collected. Selection of appropriate camera settings is crucial for collecting clear, crisp video images. The most important camera settings to consider include:
Frame rate: fast movements require high frame rates. Check out this blog post for discussion and recommendations.
Shutter speed / Exposure: should be set to allow the movement to be captured with crisp images. Faster shutter speeds or shorter exposures will capture images with less movement blur, but may reduce the image brightness.
If you are collecting high-speed movements, you may need to consider introducing additional light into your capture volume in order to capture videos that are adequately bright. In general, faster movements require higher frame rates, faster shutter speeds / shorter exposures, and more light.
Subject Attire
Theia3D provides 3D pose estimates that are robust to changes in attire. General recommendations for subject attire include:
Body-fitting clothing. Each limb should be discernible from the rest of the body.
Clothing should provide rich visual features, such as visible creasing, shadows, or other textures.
Lighting is often more important than attire color, as it impacts the visual richness of the attire. Under adequate lighting conditions black attire is acceptable, but lighter colors generally provide more visual features, especially in low-light environments.
3.2. How-To: Record Intrinsic Lens Calibrations¶
Intrinsic lens calibration trials are used to determine parameters associated with the camera lenses and are used to correct for distortion and other visual effects. Lens calibration trials are required for all camera systems that make use of a chessboard calibration method, except for Sony RX0-ii cameras. Users with wand calibration-capable systems (e.g. Qualisys Miqus, Vicon Vue or FLIR Blackfly S systems) are not required to complete lens calibration unless they are planning to use the chessboard calibration method.
Lens calibrations must be performed at least once per video resolution that will be used to record movement data, and any time the lenses or focal lengths change. If you intend on collecting data at 1080p, 720p, and 540p, you will need to perform separate lens calibration trials using each of those resolutions.
Note: Adjusting the aperture and focus of OptiTrack Prime Color cameras using the dials on their lenses does not necessitate new lens calibrations.
Recommendations:
Record lens calibration trials at a low frame rate (20-30 Hz) to reduce file size and processing time.
Keep the chessboard as flat as possible during the calibration trial.
Use a computer monitor facing the person performing the calibration to provide visual feedback during the calibration.
Move slowly and deliberately to prevent chessboard blur.
Keep the chessboard as flat as possible throughout the calibration trial.
Lens calibrations can be performed for smaller groups or individual cameras and merged afterwards. If you are finding it challenging to calibrate all at once, try recording separate groups.
Recording Lens Calibrations:
Place all cameras side by side on a desk or table, facing the same direction and capturing as similar views as possible.
Stand at a distance where the chessboard occupies ~1/4 of the camera views.
Begin the recording.
Slowly move the chessboard in a systematic grid pattern, covering the entire field of view for every camera. Ensure the chessboard goes right to every edge of every camera field of view, and every corner.
Take a step back, and repeat Step 4., covering the camera fields of view again.
While covering the field of view, angle the chessboard slightly in multiple directions, varying its orientation throughout.
When you are confident that the entire field of view has been covered for all cameras, end the recording.
To process intrinsic lens calibrations, see Lens Calibration.
3.3. How-To: Record Extrinsic Chessboard Calibrations¶
Extrinsic chessboard calibration trials are used to determine the position and orientation of every camera in your camera system relative to your desired global coordinate system. As with lens calibration trials, users with wand calibration-capable systems (e.g. Qualisys Miqus, Vicon Vue or FLIR Blackfly S systems) are not required to complete chessboard calibration. A minimum of one chessboard calibration trial should be collected every time you set up your camera system, but collecting multiple is recommended. A new calibration is required any time a camera is moved, so having multiple calibration trials collected throughout a long data collection session can help prevent data loss due to accidental or unnoticed camera movements.
Chessboard calibrations must be performed at the resolution that will be used to record your movement data. If you will be using more than one resolution during your data collection, you must record chessboard calibration trials at each resolution.
Recommendations:
Record chessboard calibration trials at a low frame rate (20-30 Hz) to reduce file size and processing time.
Keep the chessboard as flat as possible during the calibration trial.
Focus on achieving groupings of 3+ cameras that can see the chessboard at all times.
Align the chessboard with marks on the ground so you can confirm that the global coordinate system is positioned correctly when processing movement trials.
Recording Chessboard Calibrations:
Set up your camera system as desired for your data collection, following the recommendations in Data Collection Principles.
Choose or place visible marks on the ground which will be aligned with the global coordinate system, for confirmation when processing data.
Check that the chessboard is fully visible in at least 3 camera views when it is placed at the desired global coordinate system origin position.
Begin the recording.
Slowly ‘show’ the chessboard to groupings of 3 or more cameras while varying the position and orientation of the chessboard slightly. Ensure the chessboard is visible to cameras that overlap between groupings, ideally so that no fewer than 3 cameras can see the chessboard at all times. Focus on achieving groupings of 3+ cameras at all times.
When you are confident that all cameras have had sufficient views of the chessboard, slowly place the chessboard on the ground, aligned with your preselected marks.
Ensure you are not obstructing the view of the chessboard in any cameras during this localization phase.
End the recording.
To process extrinsic chessboard calibrations, see Chessboard Calibration.
3.4. How-To: Record Extrinsic Object Calibrations¶
Note: Chessboard calibration is the preferred extrinsic calibration method (rather than object calibration).
Extrinsic object calibration trials are used to determine the position and orientation of every camera in your system relative to the desired global coordinate system. Extrinsic object calibration uses the static position of a calibration object with known dimensions or known positions of specific key points on the object. These 3D dimensions or positions must be measured with high precision. The calibration object should be sufficiently large or the calibration key points should be spaced far apart within the capture volume and most points should be visible in every camera view. The key points can be coplanar or can vary in all three global dimensions.
A calibration object file must be created for your calibration object, containing 3D coordinates of the object key points. Each line of the file must contain three comma-separated values representing the global x-, y-, and z-coordinates of the key point, in millimeters. Each key point should be on its own line.
Object calibrations must be performed at the resolution that will be used to record your movement data. If you will be using more than one resolution during your data collection, you must record object calibration trials at each resolution.
Recommendations:
Record object calibrations trials at a low frame rate (20-30 Hz) to reduce file size and processing time.
Object calibration trials only need to be a few seconds long.
Recording Object Calibrations:
Place your calibration object within the capture volume, at the desired position and orientation to define the global coordinate system.
Ensure the key points on your calibration object are visible in every camera view, and there are no obstructions.
Begin the recording.
After a few seconds, end the recording.
To process extrinsic object calibrations, see Object Calibration.