Skip to main content
eScholarship
Open Access Publications from the University of California

Scalable Networked Human Daily Activity Profiling

  • Author(s): Wang, Yan Wang
  • Advisor(s): Kaiser, William J
  • et al.
Abstract

Human activity analysis is becoming increasingly important to enable preventative, diagnostic and rehabilitative measures in health and wellness applications. Wearable sensors are now taking a leading role in this area. However, daily activity profiling is still challenging due to a number of complications: 1) Method to quantify human activities are not precisely defined, especially in the domain of upper body activities. 2) Body joints have multiple degrees of freedom that are required to establish the flexibility of activity performance. Therefore, even if an agreement is achieved in activity recognition, performance of a specific activity can still present numerous variations for one person in different performance bouts; 3) Performance evaluation is also challenging without a complete set of training data since the difference between pathological movements and special personal behaviors may not be accurately classified by a general activity analysis system.

Therefore, a new solution is provided to address the above three challenges. In this dissertation, a wearable sensor based location driven high-level activity classification and 3D human motion tracking system will be presented. This system consists of three sub-systems, location detection, activity classification and motion tracking. The location detection system determines the user’s location and gives this information to the activity classification system without reliance on any other localization instrumentation or infrastructure. In addition to being a fundamental advance, this adds an additional critical benefit, the location detection system also provides navigation functionality to the end users. The activity classification system classifies the general activities the end user is performing. Since the activity set is limited by a user’s location, the classification performance will be largely improved both in terms of accuracy and efficiency. The activity classification system also indicates the high-level activity classification result. On one hand, it can be used for compliance assessment to examine whether the user is following his prescribed exercise activity according to proper protocols. On the other hand, it enables an activity specific motion tracking protocol to be used in the motion tracking system resulting in reduced analytics computation demand. The motion tracking system supports reconstruction of motions for remote and post-event visualization as well as performance evaluation based on the idea of range of motions. In the dissertation, the algorithms at the foundation of each of the three subsystems will be described as well as complete system integration.

Main Content
Current View