UCLA

A wireless health system that collects and processes data of human activities can help both users and medical professionals to monitor health status remotely. Therefore it saves tremendous medical resources and costs compared to traditional treatment in which a huge amount of human effort is involved. We present two systems that can correctly classify human daily life activities with little training, and another system to reconstruct human motion trajectories from commercial low cost MEMS inertial measurement units (IMUs) and the Microsoft® Kinect.

A system that reliably classifies daily life activities can contribute to more effective and economical treatments for patients with chronic conditions or undergoing rehabilitative therapy. We propose a universal hybrid decision tree classifier for this purpose. The tree classifier can flexibly implement different decision rules at its internal nodes, and can be adapted from a population-based model when supplemented by training data for individuals. Compared to other methods, the experimental results showed a high accuracy of classifying human daily live activities.

After we have an accurate classification of human activities, we present a system to further reconstruct motion trajectories using IMUs and the Kinect. The system fuses different motion reconstruction models to give a better tracking result, in which each model is weighted and transformed to a universal basis. This model is also expandable to accommodate different resources and environments. Experimental results showed a great improvement over past methods only using a single motion reconstruction scheme.