Design And Implementation Of Fall Detection And Protection System Based On Multi-sensor Informatio

The aging of the population is a major problem facing our country’s current economic and social development.The probability of falls in the elderly is significantly higher than other groups since the deterioration of physical functions.The health of the elderly is seriously threatened by falls.At the same time,the huge financial burden of the family will also come with the elderly fall.After the loss of balance is detected,impact injuries to the elderly can be greatly reduced by deploying an airbag,meanwhile,the elderly can get help by sending out alarm information in time.In the current fall detection study,most systems use a single sensor or signal,body movement information cannot be reflected by limited information,which limits the accuracy of the detection.Therefore,this thesis uses multi-sensor information to detect the event of fall,and develops an airbag trigger device for rapid inflation,which can be used repeatedly after changing the gas cylinder.This thesis focuses on the following research elements.(1)The overall scheme of fall detection and protection system is formulated through the analysis of software and hardware requirements,and the related technologies involved in the implementation of the scheme are introduced.The hardware data acquisition device for multi-sensor information is built,and the data used in the experiment is collected.The wearing position of the sensor is determined through analysis.Next,the hardware used in the data acquisition device is selected,and the acquisition device is set up.In order to obtain more information,the coordinate system transformation and Kalman filtering algorithm are used to calculate the information in the vertical direction.(2)By combining machine learning and mobile application,a real-time detection scheme for fall is completed.After preprocessing the collected data,two fall detection algorithms based on statistical learning and deep learning are designed.The former uses the similarity of the waveform to extract the data features,and then uses the statistical learning model to process the features to obtain whether or not to fall.The latter uses the threshold method to pre-screen the collected data,and then the convolutional neural network is used to classify the final results.By comparison,the fall detection algorithm based on deep learning is more suitable for real-time detection,and the detection accuracy and average lead time of this method can reach 99.2%and 352 ms.After that,the trained model is deployed in the developed application,which enabled real-time detection of falls before impact.(3)A reusable airbag trigger is developed,and the whole system is experimentally verified.The force is stored in advance by means of frictional self-locking.When the device is triggered,the balance state is broken by the lever structure and the electromagnet,which ingeniously realizes the inflation of the airbag.Experiments show that the average time for the device to fill the airbag is 287 ms,which is less than the lead time of 352 ms in this thesis,and fully meets the requirements of fall protection.