Reconhecimento de gestos em imagens utilizando um sensor de profundidade para o controle de um robô móvel

Abstract

A study was realized to develop a system of recognition of gesture commands applied to a mobile robot. To read the gestures was used the Kinect sensor, which allows to obtain data of the joints of a person's body and to obtain more precise data of the movements. The research was started making a study of the feasibility of the development of the system. For this, different data grouping algorithms were searched for them to be used in the research. We used K-means and Fuzzy C-means algorithms and similar results were obtained between the two algorithms, although with undefined data the Fuzzy C-means algorithm was able to work with higher efficiency. It was concluded that using the data grouping technique it would not be possible to work with gestures that had greater complexity. To solve this problem, the neural networks technique was used, which allowed a greater robustness of the system, being able to work with more complete gestures. For future work, the use of techniques such as Deep Learning is used to allow the system to achieve greater performance and to be able to treat similarities between different gestures, avoiding possible decision errors, which the system developed with neural networks of a layer faces.