More and more portable electronic equipments are released nowadays, which the developers are trying to find and design some new user-friendly handling methods. The novelty of devices with display screen, such as smartphones, is the touchscreen. It is much more dynamic, comfortable, than the traditional keyboard-based interfaces, and gives the user several new opportunities. Nevertheless, there are plenty of disadvantages of this solution. The smaller screen you have, the less accurate you hit the right button. It requires great computing capacity, and therefore the battery’s life reduces quickly. You have to be within arm’s reach of the device, and so on. The breakthrough is the now emerging hand gesture based interface, which allows you to control your devices by hand moves. Attempts and experiments have long been under way in this direction, but really only mature technologies in recent years have been. The basic motivation of the competition between different technologies is to minimize the computation requirements and to maximize the accuracy of recognition algorithms. Until now, the most effective results so far have been produced by the designers of the Gesthaar gesture recognition system, which also inspired my research.
In this thesis, I will introduce the most important accelerometer-based hand gesture recognition methods. I will present some remarkable feature extracting procedures, including Haar Wavelet Transform, and a simple feature-selection process. This procedures are used to highlight the most relevant parts and properties of the original, noisy signal, made by the accelerometer equipment. I will also specify two widespread pattern-recognizing method, the Hidden Markov Model (HMM) and the Support Vector Machine (SVM) based classifier. Then I will come to the significant design decisions in which I will describe the test environment, and the circumstances of my own gesture recognitions system’s designing. After that I will get to the implementation process and the analysis of the results of the test cases. In this part, it will be shown, that how can be reached an accuracy value above 96 percent by the cooperation of the Fast Haar Transform (FHT) and the SVM-based classifier, with appropriate parameterization, in case of 10 types of gestures.