Application Research Of Gesture Recognition In Intelligent Spinning Bike

With the continuous development of virtual reality(VR)technology,human-computer interaction has gradually entered practical application from the theoretical research stage.Gesture recognition technology,as one of the important technologies of human-computer interaction,has received widespread attention in recent years.At present,gesture recognition technology mainly obtains gesture information through monocular cameras,but the data collected by this method is difficult to accurately describe dynamic gesture characteristics.In order to improve the accuracy of gesture recognition method,this paper uses binocular stereo vision device to collect gesture data,extract multi-dimensional information in gesture action to form a feature sequence,and combine long short term memory neural network(LSTM)to classify and identify dynamic gestures.On this basis,the unity3 D platform is used to design a gesture interaction system,which is applied to the new intelligent motion bicycle products to realize the function of the user's dynamic gesture command control system.The main work content is as follows:1.A dynamic gesture data acquisition algorithm is proposed to achieve real-time tracking and sampling of hand bone points.By calculating the real-time motion rate of the finger or key bone node,the threshold judgment is used to determine the start and end points of the gesture action.Gesture data between the start and end points,as the effective data of dynamic gestures,supports the recognition and tracking of hand bone points.The craft gesture library Handicraft-Gesture is introduced to define 7 kinds of gesture actions,and the data set is made to add labels to the gesture actions through one-hot encoding.2.A sequence of features describing dynamic gestures is designed.After obtaining gesture actions with complex gesture changes,including gesture action changes,hand bending changes and gesture displacement changes,by analyzing dynamic gesture information,after extracting a single feature of each frame of data information,the distance between the fingertip and the palm,the angle of the fingertip,the height of the fingertip from the palm plane and the palm movement speed are combined into an 18-dimensional gesture feature vector,and the feature sequence can fully describe the dynamic gesture.3.A two-layer long and short-term memory neural network model based on the above dynamic gesture recognition sequence is constructed to improve the accuracy of gesture recognition.The extracted feature sequence is used as the input of the model to train the dynamic gestures including the complex posture changes and displacements of the hand.Seven customized dynamic gestures are classified and recognized on the data set,and the performance of the model is evaluated through comparative experiments.The results show that the proposed double-layer long and short-term memory neural network model has higher recognition rate than RNN network and single-layer LSTM network,and can effectively improve the recognition accuracy of dynamic gestures.4.Designed and implemented a virtual gesture recognition application based on Unity3 D platform,and applied to the intelligent motion bicycle system.In this bicycle system,the binocular stereo vision device Leap Motion is used to obtain gesture data,complete gesture recognition in the LSTM network,and adjust the bicycle resistance through the bicycle control board to simulate cycling road conditions.The test results show that the gesture recognition algorithm designed in this paper is naturally smooth and has high accuracy in practical applications.