Gesture Recognition Based On Two-dimensional MEMS Ultrasonic Phased Array

As most of the wearable devices are cumbersome and the lasers and cameras are easily affected by light, this paper puts forward a method of gesture recognition using MEMS ultrasonic array, which is not affected by light, but also avoids the complex wearing equipment.The sound field model of two-dimensional planar ultrasonic phased array has been built based on that of a single element. The influence of the size of the transmit and the receive elements, the element spacing and the number of elements on the sound field characteristics of the array are analyzed to obtain the optimal array parameters. The bidirectional radiation effect of staggered transmit and receive array is used to achieve the grating lobes suppression and optimize the element spacing and improve the array scanning range. The element radius of the transmit and the receive array are 2mm and 0.32 mm, the number of the elements are 2×4 and 15×15, the element spacings for the direction of x are11.25 mm and 7.50 mm and that of y are 15.70 mm and 7.85 mm and the elevation angle with no gate lobes range from-34 to 34 degrees.Because the information theory methods cannot be used to estimate the number of coherent sources, the spatial smoothing algorithm is used to smooth the covariance matrix of received signal matrix for coherent sources. The characteristic roots of the smoothed covariance matrix are used by the information theory method to obtain the minimum criterion value for source number estimation. Compared with the error of different information theory methods, the performance of HQ is the best. The errors of source number estimate with the change of different installation errors for element spacings were also analyzed. The installation error can be-1mm~1mm with a accurate estimation and the error of the number estimate is 1 with the installation error being-1mm~1mm and that is 0.1 without the installation error.Unitary-ESPRIT algorithm is adopted for the orientation angle estimate of the sample points. Combined with the spatial smoothing algorithm for two-dimensional array, the Unitary-ESPRIT algorithm was improved to be effective for coherent sources by using the smoothed covariance matrix ofreceived signal matrix. The errors of orientation angle estimate with the change of different installation errors for element spacings were analyzed. The installation error can be-1mm~1mm with a accurate estimation and the error of the orientation angle estimate is 1.6 degrees with the installation error being-1mm~1mm and that is 0.2 degrees without the installation error. The improved Unitary-ESPRIT algorithm and the original one were also compared and analyzed,and the improved algorithm can identify the coherent sources correctly. The performace between improved algorithm and MUSIC algorithm is compared and analyzed including the elevation and azimuth angle resolution, angle error,time-consumption. The Simulation running on Pentium Dual-Core E6300 CPU with 2G memory shows that the performance of the improved algorithm is higher than that of the MUSIC algorithm, the elevation and azimuth angle resolution are about 1 degrees and 4 degrees, the response time is about 0.6s and the maximum recognition elevation angle range from-?? to 33 degrees.A simulation method of gesture sample points estimation is designed by using the source estimation distinction and infinitesimal element to partition gesture and construct the original simulation signals. The gesture sample points are projected to the built coordinate plane by the estimate orientation angle of the sample points after source number estimation, and then the gesture sample points estimation can be realized. The gesture sample points estimation effect of different kinds of arrays, different gestures and different distances from gesture to array were compared and analyzed, the merits and demerits of different arrays and their applications were summarized.