Research And Application Of Machine Learning In Gesture Recognition

The increasing popularity of human gesture recognition technology mean that can be applied in a diverse range of field.And the wearable inertial sensor has been used widely for human activity recognition,which become a newly emerged branch of machine learning and artificial intelligence.This paper develops a novel wearable recognition device,named wristband,which consists of Microcontroller and inertial sensor.These device is capable of recognition four subtle hand actions that are less discrimination in the factory worker.We analysis the utilization in the factory,which adopt the method by using two hands together.To recognize the subtle hand action of human,the step we need to be completed include preprocess of the data,analysis of the feature and extract of the feature and so on.In the paper,specific studies as follows:(1)The hardware design of subtle hand action.In order to solve the problem that user have different habits to finish the action,such as one user finish the action by right and another finish the action by left.We suggest that wear two devices in left and right wrist.Then we observe the physical information by human hand action to determine the best joint position for collecting data.The experiment indicate that this devices is able to greatly improve the precise by the collecting data,and completely explore the data.(2)Data collection and processing of human hand action.The data of acceleration,gyroscope and attitude angle is obtained by the MPU-6050 of the device.The measure of processing the data involves eliminating interference such as flutter using filtering method,executing gradient optimization using normalization and dividing the action using the segmentation technology.The results obtained from this analysis show that the selected joint data is able to recognize the proposed hand action and improve the precision through the processing.(3)Feature analysis on different hand action.We research the feature of hand include statistics feature,physics feature and both hands feature.To deal with the feature that method is Relief-F to give the different weight for every feature.And we develop the standard for feature set,thus elevate the validity.The method can create a more efficient feature set for the classifier based on the different feature.(4)Human hand subtle action recognition based on Extreme Learning Machine.This paper clarify Extreme Learning Machine algorithms,and compare with classical algorithm.The study also provides a result that k-nearest neighbors and support vector machine algorithm has good accuracy,but there is operate the speed consumedly slowed.Meanwhile Extreme Learning Machine can be effectively accuracy of the recognition result in all test sets.It is applicable to hand subtle action recognition and many high-demand place.In the paper,the device on wrist and the algorithm with extreme learning machine is applied to recognize the hand subtle action.Experiments show that the method can improve the accuracy to recognize four hand action using two hands,and fulfill the requirements for the gesture recognition in the factory.