The following information was submitted:
Transactions: INTERNATIONAL JOURNAL of CIRCUITS, SYSTEMS and SIGNAL PROCESSING
Transactions ID Number: 20-425
Full Name: Paul Yanik
Position: Ph.D. Candidate
Age: ON
Sex: Male
Address: P.O. Box 70, Cullowhee, NC, 28723
Country: UNITED STATES
Tel:
Tel prefix:
Fax:
E-mail address: pyanik@clemson.edu
Other E-mails: pyanik@wcu.edu
Title of the Paper: Sensor placement for activity recognition: comparing video data with motion sensor data
Authors as they appear in the Paper: Paul M. Yanik, Jessica Merino, Joe Manganelli, Linnea Smolentzov, Ian D. Walker, Johnell O. Brooks, Keith E. Green
Email addresses of all the authors: pyanik@clemson.edu, jmerino@clemson.edu, jmanganelli@clemson.edu, linneas@clemson.edu, iwalker@clemson.edu, jobrook@clemson.edu, kegreen@clemson.edu
Number of paper pages: 8
Abstract: The development of ubiquitous sensing strategies in home environments underpins the promise of adaptive architectural design, assistive robotics, and services which would support a person's ability to live independently as they age. In particular, the ability to infer the actions and behavioral patterns of an individual from sensor data is key to effective design of such components for aging in place. Frequently, activity recognition is accomplished using vision based sensors. The method employed in this paper makes use of self similarities in a video motion sequence to construct a descriptor of the activity in the form of a Histogram of Oriented Gradients (HOG). Descriptors are used as exemplars for classification and are shown to accurately identify motion video recorded from other views. Three candidate motions were performed using a PUMA robot (for repeatability). Video of each motion was recorded from an array of vantage points on the surface of a virtual sp!
here surrounding the workspace of the robot. This method is then extended to non-video motion sensor data collected from the same set of points. Results show that mean HOGs generated from Self Similarity Matrices may serve as effective exemplars to classify motions in both video and non-video formats. Video data provides superior classification results. However, motion sensor data offers a less intrusive option with promising accuracy especially when multiple sensors outputs are fused to form aggregate readings.
Keywords: Activities of daily living, Activity recognition, Aging in place, Sensor placement, Self similarity
EXTENSION of the file: .pdf
Special (Invited) Session: Toward active sensor placement for activity recognition
Organizer of the Session: 650-435
How Did you learn about congress:
IP ADDRESS: 152.30.56.200
