| Abstract [eng] |
The purpose of the master's project is to create a demonstration system in which a small humanoid robot, having seen an extended hand, accepts it as a greetings and reacts to it. The visual recognition of the hand, which is the main part of the system, has proven to be a difficult task. I've described in this work how to solve this problem using deep learning. Unlike many other gesture recognition research, this work did not use depth information, video clips. The task was done using static images. For this reason, a simple camera was used, since the gesture that needs to be recognized is quite static. I have collected a special dataset for this task. Various convolutional neural network configurations and learning algorithms have been tested. However, the biggest breakthrough came when the background was removed from the images and the model was focused only on the person in front. The created demonstratior can be used as an educational tool for attracting and engaging young future researchers and specialists in exact sciences, machine learning, and robotics. |
