Topomorphological approach to automatic posture recognition in ballet dance. Issue 11 (1st November 2015)

• Record Type: Journal Article
• Authors: Saha, Sriparna; Konar, Amit
• Is Part Of: IET image processing. Volume 9:Issue 11(2015)
• Page Start: 1002
• Page End: 1011
• Publication Date: 2015-11-01
• Subjects: image segmentation -- image colour analysis -- Radon transforms -- humanities -- shape recognition -- visual databases; topomorphological approach -- automatic posture recognition -- ballet dance -- automatic unknown dance posture identification -- six stage algorithm -- skin colour segmentation -- stick figure diagrams -- minimised skeletons -- Euler number -- Radon transform -- integral plots -- fundamental postures -- initial database; Image processing -- Periodicals; 621.36705
• DOI: 10.1049/iet-ipr.2014.0622
• Languages: English
• ISSNs: 1751-9659

Abstract

Abstract : The proposed system aims at automatic identification of an unknown dance posture referring to the 20 primitive postures of ballet, simultaneously measuring the proximity of an unknown dance posture to a known primitive. A simple and novel six stage algorithm achieves the desired objective. Skin colour segmentation is performed on the dance postures, the output of which is dilated and is processed to generate skeletons of the original postures. The stick figure diagrams laden with minor irregularities are transubstantiated to generate their affirming minimised skeletons. Each of the 20 postures based on their corresponding Euler number are categorised into five groups. Simultaneously the line integral plots of the dance primitives are determined by performing Radon transform on the minimised skeletons. The line integral plots of the fundamental postures along with their Euler number populate the initial database. The group of an unknown posture is determined based on its Euler number, while successively the unknown posture's line integral plot is compared with the line integral plots of the postures belonging to that group. An empirically determined threshold finally decides on the correctness of the performed posture. While recognising unknown postures, the proposed system registers an overall accuracy of 91.35%.