In my thesis work I wrote about the implementation of a virtual keyboard application, which is controlled by the different kinds of gestures of the hand. The input device for this task was a simple web camera. First of all I described the difficulties and methods for computer vision and hand detection. In my work I presented two kinds of processes, which I have also implemented. I chose hand detection by skin color as one of the methods because the theory behind it is relatively simple. I listed two kinds of background segmentation methods: mixture of Gaussians, and the use of codebooks. After that I defined the convexity defects, so that I could get a well working model of the position and shape of the hand. The second methodology was about teaching a classifier to the recognition of the different shapes of the hand. For this goal I used the Haar-like features, which is why the process is called Haar-training. After that, I presented the concept of the integral image, and the vitals of AdaBoost algorithm. Last I showed how all of this works on a practical example. After training multiple classifiers I tried to give a good approximation to the optimal values of the parameters of the Haar-training. I introduced some tools I used to implement the application, like OpenCV and QT, with their advantages and fields of usage.