Real Time Hand detection and gesture recognition system

Abstract

Detection and recognition technique are very important in computer vision architectures. So far, it has much different object detection application on this area. These objects include face, fingerprint, human body, gesture and so on. The technique use in face detection are achieving excellent performance in real time, but those technique may not suitable for other object especially hand detection and gesture recognition. Existing hand detection approaches based on the Viola-Jones methods will be influenced by the background noise of training images and the rotation of the images. As hands are non-rigid objects, the training images always contain many other objects that decrease the performance of the training result. It then gives rise to the need of an improvement the detection method for increasing accuracy. Finger tip detection also contains lots of noise during the detection such as the shadow of the hand, the background object. It not only needs to determine the finger tip point correctly, but also need to consider how to eliminate the unnecessary point in the hand. To cope with existing needs, this study looked into both feature and learning-based technique so as to for solution of the gesture detection. The features such as skin detector and the Viola-Jones method, and a learning based algorithm, Adaptive Boosting(AdaBoost), were used for the hand detection. And also we will find out optimum method how to recognize the finger dip accurately. This project development will focus on the relationship between different factors in hand detection and gesture recognition. The different factors include the number of the training image set, stager number, and error rate and hit rate. To investigate those factor for perform better detection. On the finger dip detection side, we will focus on how to improve or replace the existing method or algorithm to detect the finger tip accurately.