Research On The Algorithm Model Of Monocular Visual Gesture Recognition Under The Skin-like Background

In recent years,gesture recognition has become an indispensable part of humancomputer interaction due to its simplicity,convenience and efficiency.However,the hardware of gesture recognition method based on data glove is expensive and inconvenient to operate,so gesture recognition based on computer vision comes into being.This research focuses on the monocular vision based gesture recognition algorithm model under the background of skin color.The main research contents include 1)static gesture recognition model and 2)dynamic gesture trajectory recognition model.The research focuses are summarized as follows:First of all,the type static hand gesture recognition model is divided into three stages: 1)gesture segmentation,because the traditional watershed algorithm in class under the background color will appear serious over-segmentation phenomenon,unable to accurately extract static hand gestures,so on the basis of the traditional watershed algorithm introduced Gaussian filtering and Hu moment centroid points,can complete static hand gestures to segment the image in the region.2)arm segmentation.Since the arm region is redundant information for static gesture recognition,this paper proposes a model combining PCA reduction and convexity detection technology,which can effectively locate the arm segmentation line and eliminate the arm region of static gestures.3)gesture recognition.A two-channel convolutional neural network is built.Experiments show that this network has a higher accuracy rate than the traditional convolutional neural network and SVM.At the same time,the static gesture recognition model proposed in this paper is compared with HOG-PCA-LBP feature fusion method to illustrate the effectiveness of the model proposed in this paper.Secondly,the dynamic gesture trajectory recognition model is divided into three stages: 1)trajectory tracking.The motion area of the gesture in the dynamic gesture video is tracked by using the three-frame difference method,and the motion area of the dynamic gesture in the skin-like background is extracted more accurately.2)track description: Hu feature moment is used to find feature points from the extracted gesture motion area and describe the gesture motion track;3)template matching.Since the traditional DTW algorithm has a large amount of computation and low operation efficiency,the retrieval method is improved based on the traditional DTW algorithm to improve the operation efficiency.Finally,it is matched with 6 dynamic gesture templates in the database to obtain the dynamic gesture trajectory classification.Experiments show that compared with the traditional DTW algorithm,the accuracy of the improved DTW algorithm is basically unchanged,but the matching efficiency is improved.Based on the above results,a real-time gesture recognition system is developed in this subject,which verifies the effectiveness of the model and proves that the system has a high accuracy and real-time performance.