Human Pose Detection And Implementation Based On Multimodal Information Fusion

Human posture recognition is widely used in today’s life.It plays an important role in medical and health protection,sports analysis,film and television production and other fields.Especially with the improvement of living standards and the aggravation of social aging,people pay more attention to their own health.Posture detection for the elderly can provide medical advice and timely help.The traditional wearable device attitude recognition uses a single sensor to collect information is relatively single,the accuracy of attitude recognition is not high,the number of sensors is large,and it is not convenient to carry.This paper designs a low-power,low-cost,lightweight multi-modal information human body attitude recognition system,which can collect,upload,recognize and store attitude data in real time.The main results are as follows:In this paper,the attitude recognition system is designed,including attitude data acquisition terminal and attitude detection platform.The data acquisition terminal is mainly composed of nine axis sensor icm20948 and main controller.The collected information is uploaded to the human posture detection platform based on pyqt5 framework through Bluetooth module hc-08.The system can recognize the posture,archive and query the data,and give an alarm in case of abnormal falling behavior.In order to improve the attitude recognition rate,the data collected by three sensors are fused to solve the attitude angle.The classical complementary filtering data fusion method has poor accuracy and low stability.According to the characteristics of complementary filtering and Kalman filtering,a method combining complementary filtering and Kalman filtering is designed in this paper.The experimental results show that the method is effective in dynamic and static state,The effect of attitude calculation is better,which can effectively suppress the drift and noise of the sensor.The data collected by the sensor are filtered and segmented.The time domain,frequency domain and attitude angle of the data are selected as the features of attitude recognition.The recognition rate is 91.4% and 90.5% respectively by using support vector machine and random forest algorithm.The results show that the attitude recognition system can realize the recognition of human posture.