Multidimensional evaluation and analysis of motion segmentation for inertial measurement unit applications

Despite various Inertial Measurement Unit (IMU) applications and their technological advances, the process of motion segmentation performed by accelerometers continues to be essential for finding the points at which motion starts and ends. In contrast to the fast growing and diverse requirements of IMU applications for motion segmentation, the evaluation of its accuracy is in need of improvement. Accuracy-oriented evaluation is unable to directly indicate motion discontinuity in the estimated results, and present enough information satisfying various requirements. To complement conventional evaluation methods, we propose a multidimensional evaluation based on new additional evaluation criteria, and justify their availability by assessing nine conventional algorithms. Through an experiment based on 462 handwriting measurements from 19 subjects, we show that algorithms with high accuracy are sensitive to movements that are unintentional and fine, but are unable to specify unexpected motion partitioning. On the other hand, we verify that our proposed metrics describe the status of both energy smoothness and parameter tuning for the motion discontinuity suppression. It also appears that a minimum time delay of 150 ms is required to reliably suppress the motion discontinuity, and algorithms with longer time delay do not always assure sufficient motion discontinuity suppression. Additionally, axial information integration performed only after securing reliable energy smoothness along each axis can guarantee significant performance improvements. As a result, it turns out that a key factor for reliable motion segmentation is how to generate well smoothed energy with minimum time delay, and the deliberate selection of algorithms with smoother energy and less time delay is a better strategy than the delicate parameter adjustment in a fixed algorithm with poorer energy status under the same condition. Using the analysis based on the proposed criteria, the selection of motion segmentation and the adjustment of the parameters for a given purpose are finally introduced as an application of the proposed multidimensional evaluation
Publisher
SPRINGER
Issue Date
2016-09
Language
ENG
Keywords

GESTURE RECOGNITION; MEMS ACCELEROMETER; NAVIGATION; PEN; MODEL

Citation

MULTIMEDIA TOOLS AND APPLICATIONS, v.75, no.18, pp.10907 - 10934

ISSN
1380-7501
DOI
10.1007/s11042-015-2812-1
URI
http://hdl.handle.net/10203/213222
Appears in Collection
ME-Journal Papers(저널논문)
Files in This Item
There are no files associated with this item.
  • Hit : 127
  • Download : 0
  • Cited 0 times in thomson ci
This item is cited by other documents in WoS
⊙ Detail Information in WoSⓡClick to seewebofscience_button

qr_code

  • mendeley

    citeulike


rss_1.0 rss_2.0 atom_1.0