3D Hand Gesture Recognition

Human gesture language is the second indispensable method in interactive communication. People can simply express and understand others’ intentions through gestures when the language sound do not work. The information technology is rapid developing, people can control the machine instruction guidance by gestures, and the interactive applications in human-computer interaction area show a booming trend. The traditional gesture recognition research in2D has matured in recent years, however, due to the uneven illumination problem, occlusion limited in2D and the development of3D scanning technology, hand gesture recognition focus on the3D scene. I also try exploratory research in3D geometry scene.3D hand gesture recognition system mainly consists as follow:data acquisition and data preprocessing, feature extraction and classification.In this paper, we use3D laser scanner to obtain3D point clouds data and establish the gesture database. After de-noising and segmentation, we capture the gesture region. We did the PC repair work by projecting transformation,3D point clouds reconstruction, distortion repair, surface reconstruction and gesture normalization.The3D data in our gesture database is static, in feature extraction step, we create the algorithm named polar coordinate rotation, compared to traditional2D feature, this spatial characteristic solve the anti-rotation invariance problem which performance well in classification with distinct feature.In gesture classification process, we classified six gestures in database by PCA+LDA in training and testing respectively, the final correct recognition reaches96.67%. This rate is higher than2D method significantly. What’s more, the speed in feature extraction and classification step can meet the real-time requirement which proved that our method is not only accuracy, robustness but also efficiency.