Research On Hand Gesture Recognition Technology Based On Machine Learning

In the era of intelligence,smart devices are gradually integrated into all aspects of people's lives.So the demand for various types of human-computer interaction technology is constantly increasing.Hand gesture is one of the common body language of people,and it is a natural,intuitive,effective and concise communication method.Hand gesture recognition based on computer vision has gradually become an important research field of human-computer interaction,and it is a hot research topic.In this thesis,the research on hand gesture recognition technology based on machine learning is deeply studied.The main work is as follows:?.Research on the multi-hand gesture image precise segmentation algorithm based on skin color: Firstly,this thesis analyzes the distribution characteristics of skin color in different color spaces under different brightness.By combining the quality of gesture segmentation in the actual scene,the most suitable one is selected.The color space is used to create a skin color model.Secondly,the multilevel hand shape features are analyzed.The minimum bound rectangle(MBR)and the minimum area bound rectangle(MABR)of the hand region are studied,and the wrist dividing line is determined,which realizes redundant removal of the arm.These work laid the foundation for the establishment of the gesture recognition model.?.Research on the hand gesture recognition algorithm based on the new RF-Net model.Firstly,this thesis describes the architecture of the RF-Net model: the BN layer,1*1 convolutional layer,dynamic learning rate,etc.are added to the AlexNet model.Based on these optimization methods,the AlexNet_I model is proposed.By fixing the AlexNet_I model in addition to the network parameters outside the fully connected layer,it is used for feature extraction of hand gesture images,that is the convolutional structure of the RF-Net model;the random forest has the advantages of strong anti-over-fitting ability and high recognition rate,and its parallel computing strategy greatly reduces the recognition duration,RF model is used to judge the hand gesture image,that is the decision structure of the RF-Net model.The RF-Net model combines the advantages of CNN's ability to extract target objects through convolution kernels and the ability of random forests to resist overfitting and high recognition rates.Through the training in Marcel dataset,HGCHA dataset and the gesture-segmented datasets,the recognition accuracy test and recognition speed test are carried out.The gesture segmentation algorithm and the gesture recognition algorithm based on RF-Net model are effective in the field of hand gesture recognition.?.Developed and implemented a hand gesture recognition control system for home scenes.Firstly,this thesis describes the software and hardware environment of the gesture recognition control system.Then,this thesis describes the hand gesture collection system and hand gesture recognition function.Secondly,the system is tested in terms of function and performance,and verified the technical advantages of the system in terms of function and performance.