The accuracy and interpretation of the joint angles produced by APDM's full body kinematics algorithms are dependent on a variety of factors. Consideration of these factors can help you collect the most accurate data and understand the limitations of the system.
The joint axes are determined by a combination of the sensor placement and calibration pose. The calibration pose is used to correct for some misalignment between the segment axes and the sensor axes. Errors in either of these can result in a misidentification of the axes of motion where some flexion instead appears as abduction, for example. This highlights the importance of the calibration pose and correct sensor placement on the subject.
Joint angles are determined by the relative orientation between adjacent segments. The individual segment orientations can drift over time in environments with poor magnetic field uniformity or internal sensor magnetization. This happens because the orientation is determined through sensor fusion with the accelerometer providing a tilt reference, the magnetometer providing a heading reference, and the gyroscope providing measurements on how the orientation changes with time. Integrating the gyroscope provides very good short-term relative orientation, but errors will accumulate over time. Additional information can be found in a companion article here that focuses on how we estimate the orientation of your Opal sensor. This highlights the importance of field calibration of the magnetometers on your sensors and attention to any significant magnetic fields in your recording environment.
Although the sensor placements were chosen to facilitate stable repeatable attachment to the body segments, some movement of the sensor relative to the bone is possible. This skin artifact will cause the orientation of the sensor to behave differently from the orientation of the segment and also introduces an error. This is mainly notable on the upper arm during abduction. This highlights that even with careful sensor placement and strapping, some level of skin artifact may be impossible to avoid.
A different issue is the representation of the joint angles. We provide Euler angles, which are a sequence of three rotations about particular axes. These three rotations are not independent, however, so when more than rotation one is large, interpretation of the compound joint angle over time becomes difficult. This shows up mainly with the shoulder angles, where the shoulder flexion, abduction, and rotation range of motion are all quite large. If the hand is pointed up directly overhead, for example, it could be described equally validly as 180 degrees of flexion or 180 degrees of abduction. This highlights that it may be difficult to make sense of the individual rotational components of the shoulder joint in particular, without looking at the compound rotation that results from these individual rotations.