UCLA

Enabling large-scale monitoring and classification of a range of motion activities is of primary importance due to the need by healthcare and fitness professionals to monitor exercises for quality and compliance. Video based motion capturing systems (e.g., VICON cameras) provide a partial solution. However, these expensive and fixed systems are not suitable for patients' at-home daily motion monitoring. Wireless motion sensors, including accelerometers and gyroscopes, can provide a low-cost, small-size, and highly-mobile option. However, acquiring robust inference of human motion trajectory via low-cost inertial sensors remains challenging. Sensor noise and drift, sensor placement errors and variation of activity over the population all lead to the necessity of a large amount of data collection. Unfortunately, such a large amount of data collection is prohibitively costly.

In observance of these issues, a series of solutions for robust human motion monitoring and activity classification will be presented. The implementation of a real-time context-guided activity classification system will be discussed. To facilitate ground truth data acquisition, we proposed a virtual inertial measurements platform to convert the currently available MoCap database into a noiseless and error-free inertial measurements database. An opportunistic calibration system which deals with sensor placement errors will be discussed. In addition, a sensor fusion approach for robust upper limb motion tracking will also be presented.