The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Download
Publications Copernicus
Download
Citation
Articles | Volume XL-5/W6
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-5/W6, 27–32, 2015
https://doi.org/10.5194/isprsarchives-XL-5-W6-27-2015
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XL-5/W6, 27–32, 2015
https://doi.org/10.5194/isprsarchives-XL-5-W6-27-2015

  18 May 2015

18 May 2015

MUSIC-ELICITED EMOTION IDENTIFICATION USING OPTICAL FLOW ANALYSIS OF HUMAN FACE

V. V. Kniaz1 and Z. N. Smirnova2 V. V. Kniaz and Z. N. Smirnova
  • 1St. Res. Institute of Aviation Systems (GOSNIIAS), Moscow, Russia
  • 2Gnessin Russian Academy of Music, Moscow, Russia

Keywords: optical flow, musical psychology, emotion identification, human face recognition

Abstract. Human emotion identification from image sequences is highly demanded nowadays. The range of possible applications can vary from an automatic smile shutter function of consumer grade digital cameras to Biofied Building technologies, which enables communication between building space and residents. The highly perceptual nature of human emotions leads to the complexity of their classification and identification. The main question arises from the subjective quality of emotional classification of events that elicit human emotions. A variety of methods for formal classification of emotions were developed in musical psychology. This work is focused on identification of human emotions evoked by musical pieces using human face tracking and optical flow analysis. Facial feature tracking algorithm used for facial feature speed and position estimation is presented.

Facial features were extracted from each image sequence using human face tracking with local binary patterns (LBP) features. Accurate relative speeds of facial features were estimated using optical flow analysis. Obtained relative positions and speeds were used as the output facial emotion vector. The algorithm was tested using original software and recorded image sequences. The proposed technique proves to give a robust identification of human emotions elicited by musical pieces. The estimated models could be used for human emotion identification from image sequences in such fields as emotion based musical background or mood dependent radio.