UMBC ebiquity
ArRf - Activity Recognition with RF
September 1, 2005 - May 1, 2006
As the population ages tools for aiding in the care of elderly become increasingly valuable. There is a need for a suite of tools that monitor senior citizens, help them through their day, and alert others if they need help. Several good techniques for creating systems that assist senior citizens have emerged. What all such computer systems lack is a good way to determine what a person is actually doing. Entering every task that a person does into a computer is time consuming and not practical. Several researchers have tried to develop systems to monitor a user�s activities and try to determine what they are doing. This is a difficult problem; most approaches have such poor performance that they are unusable. If the activities of daily life (ADL) can be accurately recognized then all of the previous work involving planning, care giver notification, etc can be realized. This experiment will identify the activities of daily living (ADL) through data collected from RFID tags. This work will build upon the work of Philipose et al.[1] That team created a RFID tag reader built into a glove. They then put RFID tags on many items in the house. In this way they are able to determine what items the person touched. To understand the stream of RFID tags encountered by the reader the team used a combination of natural language processing and a Bayesian network where the probabilities were derived from the web via the Google API. In this experiment the ADLs will be modeled using either XML or RDF. The stream of RFID data will be interpreted by a trained classifier. I will try a variety of different classifiers: Hidden Markov Models (HMMs), neural networks, learning bayes networks, genetic algorithms or learned decision trees. This training could create a personal ADL recognition system. The customization may improve the accuracy of an ADL recognition system.