Optical motion capture technologies were first used in biomechanics research studies in the late 1970s and early 1980s. Current biomechanical applications include gait analysis, ergonomics or human factors studies, orthopaedic evaluations, and a wide range of sports performance studies. Optical motion capture technology is also used quite extensively for computer animation work for video games, television shows, and Hollywood movies.
A typical optical motion capture system will consist of at least 3 up to 48 or more cameras in combination with a computer incorporating system controller software to automate the data collection. The cameras and computer controller are what makes optical motion capture systems more expensive than any other alternatives. The number of cameras required for an application is dependent upon the number of subjects being recorded as well as the desired capture area. The more subjects and/or the larger the capture area, the more cameras that will be needed for the laboratory or studio. This is independent of the type of markers used for the system.There are two different types of markers used with optical motion capture systems. Passive markers are circular or spherical markers coated with a retro-reflective material to reflect light. Cameras in a passive marker system are equipped with light emitting diodes (LED) and the light emitted by the LEDs are reflected by the markers. The passive camera’s threshold settings can be adjusted so that only the bright reflective markers are captured and all other reflections from skin and clothing are ignored.
In an active optical system, the markers themselves are LEDs. Some active marker systems illuminate 1 LED at a time (albeit very quickly), thereby eliminating the need for identifying each marker. Other systems illuminate all LEDs at once. In these systems, the amplitude or frequency of each marker is modulated to identify each marker. In either case, the system controller knows which marker is which at all times and therefore does not require any post-processing marker cleanup to identify markers after data occlusions. Because the LEDs themselves are powered on during the session, they require cables run to the markers. Most systems have tethered capability where a control box can be attached to the subject and the active markers connected to the box, which can be either connected to computer controller through a cable or some systems offer wireless connection. Below is an example of a curved rigid body plate with 6 markers used for an NDI Optotrak system that we used in the Callaway Golf Player Performance Center.
- Marker Placement – either active or reflective markers are placed on the subject according to the marker placement protocol used. Marker placement is a very important step in optical motion capture. See the marker placement protocols for a more thorough discussion.
- Camera calibration – Every camera in the mocap session records each marker as a global 2D position. At least 2 different cameras need to see a marker in order to calculate the global 3D position of that marker. However, before calculating 3D data, the cameras need to be calibrated. The camera calibration process is used to find the exact position and orientation of the camera in the global space, as well as the camera internal aspects, like focal length and image sensor position. All of these parameters allow one to define a mathematical correspondence between the coordinates from the image plane (given in pixels) and the global coordinates (given in a unit of length). Direct Linear Transformation (DLT) approach is the most common calibration technique and is relatively simple and provides good precision for 3D calculation when working with small reconstruction volumes. Most mocap systems have an automatic calibration routine included in their control software which requires a dynamic calibration where a calibration wand or cube is waved through the capture area for some small time frame. DLT calibration wands or cubes have markers permanently attached at well-known positions. When calibrating the system, the calibration tool must be briefly positioned in the capture volume where the movements will be taken so that each camera can register at least one image of it. To maximize accuracy and precision, the calibration tool is moved around the entire capture area volume including rotations for a dynamic calibration. Another trial is usually recorded with the calibration wand/cube positioned in order to define the global origin and XYZ axes.
- Capture – after the cameras are calibrated, the global origin defined, and the marker placement protocol has been applied to the subject, then motion data is captured for the desired study. This will vary significantly based on the study being performed. It is important that the markers are attached sufficiently such that they don’t move from trial to trial and should be positioned the same from the first to last trial.
- 2D Marker identification & 3D marker reconstruction – this step is usually applied automatically in the motion capture control software. The center of each marker is estimated as a position within the two-dimensional (2D) image that is captured for each camera. The control software typically just has to be pointed to the previously recorded calibration trial, and the software will automatically take the 2D marker identification from each camera and reconstruct the 3D position for each marker in terms of global coordinates in the XYZ reference frame.
- Cleanup – both passive and active optical markers are very susceptible to data occlusion during a mocap trial. Markers have to be seen by the camera at each sample in order to calculate the 3D position of the marker. If a marker is not seen by at least 2 cameras, then it is not possible to calculate the global 3D position. Occlusions occur if body rotations cause a marker to be blocked from 1 or more cameras. They also can occur if multiple subjects are being recorded and some of the markers are blocked. Occlusions are a bigger problem for passive markers; active markers know the number of each marker as they are strobed by the controller once for each sample. So even though an active marker may be occluded for one or more time samples, the controller will know what each marker is if a 2D position is recorded by each camera. In the case of passive markers, the controller does not know what each marker is and misidentified data or marker swap can occur if there are a number of markers close to each other at a time sample or when markers are occluded for a sample and then come back. So when data occlusion occurs with passive markers, it is a much bigger problem as data handlers need to manually go in and cleanup the data to handle the marker swap or missing data. There are automatic proximity routines that can be used to reconstruct the 3D data instead of having to manually do it. However, it still often requires the data handler to examine the data after the capture session as they typically have to filter or interpolate the noisy data anyways. When too many markers are occluded or the duration of an occlusion is too long, it is impossible to fix the problem. That is one of the reasons why passive optical systems usually use an excessive number of cameras for redundancy so that they don’t have to deal with any data occlusions.
- Subject calibration – one additional step may be required dependent upon what the output mocap data will be used for. Often times it is necessary to calibrate (or scale) the subject for character animation or if it will be used for subsequent inverse dynamics routines. Subject calibration involves identifying where the markers are in relation to their joint centers or other bony landmarks. This usually involves capturing a trial with the subject in a T pose with their arms extended 90 degrees straight out sideways (cross position). This helps define a known orientation with all markers in view to the cameras to define a 3D position for each marker in this pose. More detailed analyses require digitizing bony landmarks relative to the optical markers placed on the subject in order to define joint centers relative to the measured 3D marker positions.
As with the other motion tracking technologies, there are advantages and disadvantages to optical tracking technologies.
Optical Tracking Systems Advantages
- Data output from a state of the art optical system is very clean and accurate when it does not suffer from occlusion problems.
- Capture rates are high (typically 120 Hz up to 1000 Hz dependent upon the configuration used)
- Multiple subjects can be captured simultaneously given enough cameras to minimize data occlusions.
- A large number of markers can be used given enough cameras to minimize data occlusions.
- Marker configurations can be changed easily, dependent upon the program goals.
- Optical systems’ capture subjects can move freely in the capture volume.
- The capture volume is only limited by laboratory space and the number of cameras used in the capture project.
Optical Tracking Systems Disadvantages
- Markers can be occluded by capture subject movement, other subjects, or props resulting in loss of data. This is a much bigger problem with passive markers.
- Extensive post-processing may be necessary to handle marker swap, missing data, and noisy data.
- Rotational data needs to be computed from positional data in post-processing.
- Lighting needs to be controlled during the session in order to insure adequate reflection of the markers, especially for passive markers.
- Real-time visual feedback is typically limited to stick figures if using the optical systems’ software program.
- The cost of the hardware is typically much higher than other technologies; if the project involves multiple subjects, a large degree of upper extremity rotation during the motion, or a large capture volume, then this problem becomes even bigger as more cameras are necessary to minimize data occlusions.