Research Seminar - September 20, 2002

Recognising Moving Handshapes

Abstract:

The automatic recognition of hand shapes in a moving image sequence requires the elements of hand tracking, feature extraction and classification. In this seminar, I will present a new shape feature extraction technique. The proposed technique tracks the hand throughout an image sequence using a combination of pattern matching and condensation algorithms. Hand Region Of Interest (ROI) is converted into a polar image where finger-only regions are automatically detected and are used for feature extraction. A 2D hand shape is represented by a topological formation of the finger contour. Euclidean distances between finger contour and centre of the hand are extracted as a 1D signal sequence. The signal sequence is represented by cepstral coefficients that are often used in speech recognition. These 1D features are used to form a feature set to represent a 2D hand shape. Hand shapes in a moving sequence are then classified, using Mahalanobis distances from the known hand shape models. The method and experimental results will be presented at the seminar.