Application Of Sensor Information Integration In Human Motion Capture And Recognition

Human motion posture recognition combined with multi sensors informationbelongs to the research direction in the field of pattern recognition. It includes aprocess of data collection using a variety of motion-sensative sensors, datatransimission, noise suppression, and classification based on feature extraction.Comparing with human motion posture recognition technology based oncomputer vision technology, the technology of human motion posture recognitionand monitoring based on multi sensors information has the advantages of privacyprotection, portablity and accuracy.The data acquisition unit used in the paper mainly includes accelerationsensor, angular velocity sensor and MCU. The unit analyzes the data by themeans of time domain and frequency domain respectively. After that, Bayesclassifier is used for classifying and recognizing the human current motionposture. The main contents in this paper include the following aspects:1. The design of the hardware platformThe human motion posture recognition and acquisition unit, which is basedon two kinds of sensor, is a human motion acquuisition terminal. The humanmotion posture recognition system is made up of microprocessor, three-axisacceleration sensor, three-axis angular velocity sensor and a central control unit.The embedded software is developed using Keil uVision4developmentenvironment. C language is used to design the MCU program in the compiler ofKeil uVision4from top to bottom.2. Acquisition and pre-processing of the human motion dataThe MCU controls the sensors to collect the human motion posture data, andthen the data is saved to the64K Flash of IAP15F2K61S2microcontrollerdirectly. And then the MCU uploads the data to the PC for data preprocessing through the UART protocol.3. Data analysis based on time-domain and frequency-domain methodThe body posture can be quickly recognized using acceleration model valuein the time domain. However, the robustness is very low which means wrongclassification may occure. In order to reduce this recognition error, we combinethe time-domain method with the frequency-domain characteristics to analyzesome daily body posture (especially the posture fall of body). This approachimproves the correct rate on identifying the body posture of the current campaign.4. Posture recognition of upstairs and downstairs based on Bayes classifierThe Bayes classifier is used for recognizing the posture of upstairs anddownstairs. The data collected by the acceleration sensor and gyroscope isfiltered and calculated firstly in the method. After featur extraction andintegration of acceleration and angular velocity data, the Bayes classifier is usedto identify upstair and downstair movement. Experiments show the method caneffectively distinguish the posture of upstairs and downstairs.