IPARS: Intelligent Portable Activity Recognition System
Date Issued
2006
Date
2006
Author(s)
Lin, Chi-Yau
DOI
en-US
Abstract
The things we normally do in daily living including any daily activity we perform
for self-care (such as feeding ourselves, bathing, dressing, grooming), work,
homemaking, and leisure. The ability or inability to perform activities of daily living
(ADLs) can be used as a very practical measure of ability/disability in many
disorders. Modeling human ADLs via contextual information is gaining increasing
interest in the artificial intelligent and ubiquitous communities.
There are several studies on tracking ADLs, such as using video cameras, or microphones.
Many people are uncomfortable living with cameras and microphones.
Furthermore, video cameras, especially in non-public place spaces, provoke strong
privacy concerns. Those approaches sometimes cannot process sensor data with
minimal computational resources (e.g., a personal digital assistant (PDA)). Our
developed system: Intelligent Portable Activity Recognition System (IPARS) performs
activity recognition online with minimal computational resources. Sensors
should be low-maintenance, easy to replace and maintain. Tagging objects with a
remotely readable identification tag is adopted in our system. In addition, we develop
a wearable wrist, based on Radio Frequency Identification (RFID) reader to
detect everyday objects. In addition, we use WiFi positioning system to capture a
person’s current position. IPARS is equipped with an RFID reader, which connects
to a PDA. The way to obtain contexts to infer the current activity of a person is by
detecting person-object interactions, and movement.
Our approach uses a general framework for activity recognition by building upon
and extending multiway tree structure (trie) to model ADLs via contextual information.
There are two steps for IPARS to achieve activity recognition.
First, by using the interface provided by IPARS, the person can train his activities
easily. Therefore, while the person performs activities, IPARS models sequential
sensor readings involved in these activities. Second, after modeling human activities,
IPARS makes inferences for activity recognition by collecting current contexts
and extracting features to map trained activity models.
There are two phases in our experiments. In the first phase, we conducted
our experiments in the Computer Science and Information Engineering at National
Taiwan University. The first goal is using IPARS to test object recognition and
location tracking. In the second phase, the experiment is run in a real home. The
second goal is to evaluate the proposed solution of the activity recognition problem.
We found that a discriminative relational approach for activity recognition based on
the framework of multi-tries models to be well-suited to model sequence of contexts
for activity recognition. IPARS detected 80 percent correctly for activity recognition.
The results are promising. In the future, we focus on how to detect activities about
healthcare. We plan to extend our model in a number of ways. First, by collecting
data from more subjects, we can learn a set of generic models by clustering the
subjects based on their similarities; then we can use a mixture of these models to
better recognize activities of a new person.
Subjects
動作辨識
可攜式系統
可穿戴式感應器
定位追蹤
Activity recognition
RFID
ADL
WiFi
LeZi Trie
SDGs
Type
thesis
File(s)![Thumbnail Image]()
Loading...
Name
ntu-95-R93922129-1.pdf
Size
23.31 KB
Format
Adobe PDF
Checksum
(MD5):9d865506aa6ec528be8d560edbb51fb3