Abstract:
Natural user interface is one of the major focuses of research in computer vision because of computers enlarge integration in our daily lives. For intuitive human-computer interaction, hand gestures are the appealing option rather than face or eye gaze. To achieve this goal, computer should be made responsive and visually intelligent so it can recognize hand gestures. Hand recognition is complex and challenging problem due to its high degree of freedom.
Many preceding research shows the detection based on edge, texture, and color to classify an object. Those techniques were less efficient. This thesis show a low-level approach which focus on hand detection and tracking with Haar-like features and AdaBoost algorithm. The Haar-like features are used to get the appearance properties of objects. The adaboost algorithm speeds up the training and constructs accurate strong classifier by combining weak classifiers. Different hand postures are recognized by parallel cascading. The acquired coordinates are used to generate touch events in different computer screens which virtually create a touchless interface. With implementation of this system, virtual objects can be manipulated by different hand movements. An application of this robust identification system is that different inputs can be generated by different postures.
