Research On Air Handwriting Recognition Algorithm Based On Wearable Sensor

Due to the rapid development of virtual reality,computer and environment interaction technology,aerial handwriting biometric technology plays an increasingly important role in human-computer interaction,becoming a bridge between the real world and the virtual world.Air handwriting recognition generally has two recognition methods: vision-based and sensor-based.Because vision-based handwriting recognition has higher requirements for the environment and external equipment,at the same time,with the rapid development of micro electro mechanical system(MEMS)sensing technology,sensor-based air Handwriting recognition authentication has become a popular technology today.This paper uses MEMS sensors to study air handwriting recognition technology.MEMS sensors collect sensor information and cannot directly display the trajectory information of handwriting in the air.Therefore,handwriting recognition algorithms based on sensors generally process data such as acceleration or angular velocity according to specified actions,and recognize handwriting according to the characteristics of each specified action.Trajectory.This method requires manual intervention,and the user cannot perform handwriting flexibly.For the interactive control needs of the virtual scene and to make the writing movement more flexible,this paper proposes an aerial writing method based on wearable sensors.Through spatial positioning technology,the sensor data is transformed into data in three-dimensional space,record the coordinate points of handwriting movement in space.Due to the poor accuracy and drift of the accelerometer and gyroscope,the positioning error is caused.In order to improve the accuracy of air handwriting recognition,this paper proposes an attitude angle and acceleration error correction algorithm based on adaptive Kalman filter.The algorithm integrates the PI control algorithm based on the Kalman filter.The measurement results of the gyroscope and accelerometer are fused through the adaptive Kalman filter.At the same time,the PI controller adjusts the observed noise covariance matrix in real time,eliminating the noise prediction The inaccuracy results in filtering divergence,and high-precision attitude angle data and acceleration data are obtained,and then the acceleration data is integrated to realize positioning.Experimental results show that the algorithm can eliminate errors while ensuring the accuracy and robustness of the algorithm.After the handwritten space positioning data is obtained by integration,the data signal is firstly reduced in dimensionality,and then input into the support vector machine classifier for training and recognition.In order to extract more accurate feature information,this paper first normalizes the data,then extracts the time distance feature and the statistical similarity feature to obtain the feature vector,and then uses the support vector machine for training and classification recognition.Experimental results show that the spatial data converted from acceleration data and angular velocity data can accurately record the trajectory of air handwriting in three-dimensional space,which proves that the direct acquisition of sensor data into spatial data and classification and recognition can achieve better classification results.Through the support vector machine algorithm to train and recognize the air handwritten data,the training speed and the accuracy of classification and recognition have been improved,and the classification recognition rate of 92.8% is obtained,which is much higher than the recognition rate of the ordinary support vector machine algorithm.The DTW algorithm has also been greatly improved.