Multi-touch Gestures Recognition Using the Strategy of Graph Embedding
Résumé
In this master thesis, we propose a new system for multi-touch recognition. We consider a multi-touch gesture a gesture which can be performed on a multi-touch device like a tablet using more than one hand at the same time. A multi-touch gesture is rich in information. We can record from such gesture the position of each finger and its movement as well as the synchronization between the different fingers movements involved in same gestures. To deal with the complexity of these kind of gestures, we propose to model them using graphs. A graph as a general data structure allows us to represent all the relations that may exist between the graph components. We propose to represent a gesture primitive by a set of labeled nodes in the graph and the relations, either temporal or spatial, that may exist between the fingers movements by labeled edges. Graph matching is required for gestures discrimination. This task is a complex problem, to overcome this difficulty, we will explore an approach which is known by the graph embedding. This approach involves the mapping of a graph into a one dimensional vector. The advantage of this strategy is the availability of statistical tools which allow us to easily compare vectors. The results obtained at the end prove that this strategy does suite the purpose of multi-touch recognition problem.