Title: Realism in Activity Recognition for Long Term Sensor Network Deployments

Automatic activity recognition using wearables and in-situ sensors has the potential for improving both smart home health care and general services found in smart cities. However, useable solutions must account for realisms found in everyday life and be able to react as conditions change over the long lifetimes of the use of the devices. To understand the details of the required realisms and to show solutions work in these conditions, more than 17 real deployments, some as long as 9 months, were conducted. Overall evaluations used both public data sets and these real deployments.

For smart home health care, a system called DAVE was created to support the monitoring of dementia patients. This system is a comprehensive set of acoustic event detection techniques to detect 5 important verbal agitations: asking for help, making verbal sexual advances, asking questions, cursing, and talking with repetitive sentences. The intellectual merit of DAVE includes combining acoustic signal processing with three different text mining paradigms to detect verbal events. Another issue for dementia patients living alone is safety. In order to notify patients or their children about potentially unsafe situations and to track mistakes or efficiency in performing activities, it is important to monitor the quality of performing an activity and identify the missing/ wrong steps. However, the state-of-the-art activity recognition frameworks typically ignore such details and impose constraints on the types of detected activities (e.g., they might not address parallel/interleaved/joint activities), or the number of users, which reduce the robustness of the system in real world settings. To solve this problem, QuActive was developed. QuActive is a novel grammar based general purpose framework for modeling activities and micro-activities that retains the details of the activity steps, quantifies activity quality, and notifies users about missing steps and unsafe situations. For a different smart health problem, caregivers' compliance with hand hygiene is one of the most effective tools in preventing healthcare associated infections (HAIs) in hospitals and clinics. A smart watch based solution was created to detect failure to wash hands and to remind the person to wash when they forget. Broad impacts of this smart health work include improved support for dementia patients and to reduce infections due to hand washing non-conformance.

As smart cities increase their services many day-to-day realisms must also be addressed. However, as the complexities of services grow there is an increasing potential for conflicts within and across smart city services. These conflicts can cause unsafe situations and disrupt the benefits that the services were originally intended to provide. As a major intellectual contribution, CityGuard, was conceived, designed, and evaluated as a safety aware watchdog architecture. It carefully defines and enumerates the main issues regarding the detection and resolution of conflicts across services in smart cities. The broad impact for these smart city results includes that it can help prevent various city safety hazards and help support reducing pollution and traffic congestion.

Last Modified: 09/05/2017
Modified by: John A Stankovic