Design and implementation of real-time software for sourceless full body tracking using small inertial/magnetic sensors.

摘要

Practical full body tracking remains a difficult challenge related to the construction of immersive synthetic environment systems. Human body posture can be tracked in real-time using small, inexpensive inertial/magnetic sensor modules to track the orientation of individual limb segments and using the individual orientations to determine body posture. If the position of one point on the body is tracked, the position of all body segments can be determined. The software required for such a system is inherently complex due to its real-time distributed nature and the necessity to interface with varying sensor types and numbers. Furthermore, the software must be modular and flexible to ease enhancements and modifications.;This thesis describes the implementation and design of a software system capable of processing data from a minimum of fifteen inertial/magnetic sensors at an update rate in excess of 100 Hz. Sensor data can be collected by a wearable computer and submitted via wireless LAN to a fixed workstation. The workstation processes data and acts as a server for avatar animation data. Design of the system software was completed using object-oriented techniques and the unified modeling language (UML). Several conversion classes process raw data from the fifteen sensors before being processed by a quaternion based complementary filtering algorithm to produce orientation estimates in quaternion form. Estimates are sent to a graphical display unit through a local area network. At any point through the conversion process, data may be archived for data analysis purposes or reprocessing.