The Hand Subtle Action Recognition Based On Wearable Device

With the rapid development of sensor technology,microchip and art ificial intelligence theory,human behavior recognition based on wearable devices has attracted wide attention,in recent years,the smart hardware technology has developed vigorously,and the traditional external wearable sensor has turned into an intelligent wearable mobile device.Mobile devices have powerful perception and comput ing power.It can obtain and calculate the sensor data of human behavior in real time,which can provide more convenient channels for recognizing the fine motion of human hands.Therefore,how to use intelligent wearable mobile device to recognize human be havior has become a new research hotspot.The main purpose of this thesis is to use the sensor raw data generated by the human hand movement.After analysis and processing,the combination of pattern recognition technology can be used to perceive and recognize the hand subtle action.By analyzing the human hand movements and summarizing the state of hand movements,this paper divides the hand movements into three types of hand movements: arm,wrist and finger.In addition,this thesis adopts the smartwatch which is commercially available commercially,and tries to distinguish the three types of hand movements and the detailed action categories.Through the use of COTS intelligent mobile devices,this thesis provides a feasible reference scheme to realize the recognition of fine motion in human hands.There are two main parts of this study: the extraction of action segment and hierarchical recognition of the hand subtle action.In the first part,the data is collected and preprocessed by the micro movement data of the enemy department,and an improved algorithm AMFEIA is proposed,the algorithm can detect the real starting point and the ending point of a single action so that the system can recognize the human action actively.The second part,this thesis proposes a human hand motion recognition method based on hierarchical,the first layer through the calculation of motion energy for the classification of the arm,wris,finger gestures.The second level training is based on the recognition model of detailed movements under different hand movements.In order to enhance the robustness and accuracy of arm motion recognition,this paper designed a template library matching algorithm based on acceleration zero and first wave positive and negative to identify it.At the same time,considering the weak and delicate characteristics of wrist and finger movement data,this paper uses the decision tree classification model to recognize this kind of detailed action.Through two classification structures,the overall recognition scheme for the fine motion of the human hand was constructed,thus obtaining the final recognition result of the small hand movements.In this research,2000 sets of hand motion data feature vector datasets were constructed to verify the algorithm and scheme of the thesis.The experimental results showed that the accuracy rate of the human hand movements(arm,wrist,finger)reaches 100 %,in which the average accurate recognition rate of detailed movements of arm,wrist and finger reaches 97.27%,97% and 86.2% respectively.In the actual project,we used smartwatch for Android APP development and implemented the code for this identification scheme.