An ICA based head movement classification system using video signals

Abstract

Automatic head movement detection and analysis have significant applications in automobile safety systems, sign language and human computer interaction. Present systems employ algorithms that are based on various states of head, in the movement. These state based systems perform well in many cases, but lacks the ability for classification of the trajectory of head movement. Thus, in this paper, we introduce a new real-time head movement classification system from video streams based on the trajectory. This system is based on facial landmark coordinates and independent component analysis (ICA), which is a relatively new signal processing tool that reveals hidden patterns in the data. We have used a single front camera to capture the video streams and employ an active appearance model based system for finding the facial landmark points. Then ICA is carried out on these landmark points/coordinates, resulting in independent components (ICs). These IC’s are compared with reference IC’s and similarity values obtained were used for classification of the head movement by the neural network classifier. Our system achieved \(98\%\) classification accuracy and it is capable of precise classification of head movements such as left, right, up and down. A comparative study shows that the proposed system out performs all the existing systems, in terms of higher classification accuracy (\(98\%\)) and has an excellent real-time performance in terms of average time taken for detection and classification (0.085 s).