Recognition of hand gestures with kinect as commands

The aim of this thesis is hand gesture recognition (HGR) with MS Kinect as commands. For his aim, some topics such as sign language, systems with/without Kinect were examined. During the studies, advantage/disadvantage features of Kinect and the differences of different type of cameras were learned. The important clues related with subject and results are given below. For hand gesture recognition, totally 24 hand gestures were used. 18 of them are inclusive of American Sign Language (ASL). For each hand gesture, 100 images were recorded using Kinect. 100 images were taken from three persons for increasing accuracy of the proposed system. So, totally 24000 images were used in this study. The offline images were recorded and then RGB and depth information of each image were extracted. The last feature extraction method of histogram of oriented gradients (HOG) was applied for each image. These features are used for hand gesture recognition. Three classification methods such as support vector machines (SVM), multilayer perceptron artificial neural networks (ANN) and k-nearest neighbors algorithm (kNN) were employed. After the training stage, the proposed system having three different classifiers tested in real time. 10 gestures per character were tested online and the each result such as true/false was recorded. The average of performance of kNN, SVM and ANN are 92.5%, 95.83% and 97.50% respectively. These obtained results are acceptable level, if we compare the results with other related studies.