| In recent years,with the development of science and technology,new types of human-computer interaction have emerged.Traditional contact methods such as mouse and keyboard have been unable to meet the requirements of users.As the most vivid and natural way of information communication,gestures have gradually been applied to the field of human-computer interaction.This paper is mainly based on computer vision for naked hand gesture recognition without the aid of other contact-type auxiliary devices.The main work contents are as follows:First,traditional gesture recognition technologies generally include gesture image segmentation,gesture feature extraction,and gesture classification recognition.Skin color is widely used in gesture segmentation as important information for human hands.However,in traditional gesture recognition systems based on skin color segmentation and HU moment features,factors such as illumination intensity,experimental background complexity,and skin color-like regions in the video all have an impact on the accuracy of gesture recognition.A gesture recognition method based on SIFT feature extraction algorithm and multi-features.In view of the fact that multiple gesture targets may appear in the actual situation and the same operator may also have problems with both hands occlusion,this paper proposes to use the gradient direction histogram HOG combined with principal component analysis PCA to identify the left and right hands,and compare the paper in different environments.The accuracy of the proposed multi-feature fusion gesture recognition method can be seen through comparison experiments.The proposed method achieves a good recognition effect.Secondly,the traditional gesture recognition method needs to manually extract features and then design the classifier.The process is complex and the recognition accuracy is greatly affected by other factors.The convolutional neural network image recognition method trains an end-to-end convolutional neural network to train the input image and simultaneously extract and classify the features.Without other steps,the classification result can be directly Make the output.This paper improves the classical convolutional neural network model and builds a new model to train the model.Because the experimental data set is small,the model is optimized by introducing a 2-layer dropout layer,which effectively alleviates the over-fitting problem caused by the lack of samples.For the gesture recognition,it is necessary to call the camera to capture the effect of the illumination change when the gesture is captured.The processing stage performs binarization processing and background mask processing to overcome the influence of environmental changes,and compares the accuracy of the model on the test set under different parameters to determine the optimal parameters for the network model initialization.Compared with the traditional gesture recognition method,the recognition based on convolutional neural network has obvious improvement in both recognition accuracy and system robustness.Finally,based on the CNN model built in this paper,a set of gesture interaction scheme based on computer vision is proposed.Through gesture recognition,the contactless control image is enlarged,reduced,translated and rotated.Aiming at the inefficiency in the process of gesture recognition and the high error rate of identifying user intent,this paper proposes an algorithm for extracting key frames of video and dynamically processes the threshold to ensure high efficiency and high accuracy of the interactive system.Experiments show that the method achieves the expected effect in practical applications and achieves the purpose of gesture control human-computer interaction. |
PDF Full Text Request