Pattern mining from saccadic motion data

Peter Liang, Yingzhen Yang, Yang Cai

Research output: Contribution to journalArticlepeer-review

3 Scopus citations

Abstract

A saccade contains fixations between rapid movements. Human movements are often saccadic in a fast forwarding video tape. In this paper, we present a novel model for pattern representation and pattern matching in saccadic motions, by converting a two-dimensional saccadic motion sequence into a string of letters or numbers, with linear, extended chain code or direct encoding methods. This enables us to cluster and pattern matching with the fast text search algorithm. Our model is tested with the data of eye movement in video analysis and human movement in a building. The results show that the both extended chain code and linear encoding methods can be applied to eye gazing data analysis effectively. Extended chain code yields more accuracy in pattern clustering. However, it may accumulate errors when the motion pattern sequence is long and contain parallel subsequences. The direct labeling method works effectively in the Smart Building data analysis. Using the fast text algorithm, we found interesting patterns of the human movement.

Original languageEnglish (US)
Pages (from-to)2539-2547
Number of pages9
JournalProcedia Computer Science
Volume1
Issue number1
DOIs
StatePublished - 2010
Externally publishedYes

Keywords

  • Clustering
  • Data mining
  • Encoding
  • Saccadic motion
  • Sensor network
  • Smart environment

ASJC Scopus subject areas

  • Computer Science(all)

Fingerprint

Dive into the research topics of 'Pattern mining from saccadic motion data'. Together they form a unique fingerprint.

Cite this