Design And Implementation Of Position Detection System Based On Machine Learning

With of the economic development and transition in China,occupations with manual labor become less while indoor work and study take up a growing amount of daily time.Consequently,the percentage of sedentary population is increasing yearly.When sitting for a long time,people can hardly keep sitting in a healthy way.Unhealthy sitting for long periods may increase the risk of various diseases in eyes,neck vertebra acantha and lumbar.Therefore,it is of great significance to develop a sitting posture monitoring system to help people keep a healthy posture of sitting,form a good sitting habit and prevent physical diseases.Currently,there are two popular methods for identifying sitting postures in the domestic and overseas.One is based on video surveillance analyze technology,which is to use camera and record images of user's sitting postures,process images to get feature data and then identify different sitting postures by employing various algorithms.The other way is to use sensors and get the posture information of human body and achieve the identification.In the present paper,flexible pressure sensors which are laid on a chair are used to obtain the pressure distribution from human body.This monitoring method has a relative wide application scope and will not bring extra inconvenience to users.The sitting posture identification and classification system designing includes hardware part and software part in this paper.The hardware part is based on array distributed flexible resistive pressure sensor and adopt microcontroller chip(STM32F103VCT6,STMicroelectronics)as the controller.Signal measurement and acquisition circuit is composed of decoupling circuit,multi-channel analog switch circuit and signal measurement and acquisition circuit.SD card is used here to store data through serial communication with SCM and Bluetooth module is employed to send data to the Android system.The software part contains modules of data acquisition,data processing,data storage and data sharing.The algorithms used for sitting posture identification and classification are Machine learning algorithms.Three traditional Machine learning algorithms which including Vectors Machine(SVM),Random Forest(RF),Artificial Neural Networks(ANNs)and one Deep learning,listed as Convolutional Neural Networks(CNNs),are trained on PC platform.After comparing the identification accuracy rate of SVM,RM,ANNs and CNNs algorithms,the most capable one,CNNs algorithm with automatic feature extraction,is selected to develop classifier model and the code is finally ported to display on Android mobile terminal.