Research On Fall-oriented Support Vector Machine Detection Algorithm And Airbag Protection System

With the aging of the human body and the decline of various functions of the body,the risk of injuries from falls is gradually increasing.Falling has increasingly become a key factor threatening the physical and mental health of the elderly in our country.It not only affects the ability of the elderly to live independently,but also seriously affects the elderly.His life is in danger.Therefore,it is of great practical significance to develop and design a system that detects the activity status of the elderly in real time and provides safety protection for the elderly in the event of a fall and reduces the risk of injury caused by the fall of the elderly.Through the analysis of common fall detection schemes at home and abroad,the current technology used for fall detection roughly consists of three parts:fall detection based on image/video stream,fall detection based on human surroundings,and fall detection based on sensor/microcontroller.Although the fall detection based on the image/video stream and the fall detection based on the surrounding environment of the human body have a high recognition accuracy,the privacy of the user cannot be effectively guaranteed during the data collection process,and the anti-interference ability is weak,the use cost is high,and it is applicable the range is small.The fall detection based on the sensor/microcontroller not only meets the requirements of portability and protection of user privacy,but also has the advantages of small impact on the daily activities of the human body,wide application range,and strong scalability.At present,the most widely used sensor/microcontroller-based fall detection is to use the data collected by the sensor to set appropriate thresholds for falls and daily activities to achieve fall discrimination.This type of technology has low recognition accuracy,sensitivity,and specificity,and the threshold choice is highly subjective.Most of the existing fall protection systems use the steering gear to push the puncture needle to pierce the gas cylinder to inflate the airbag.This solution has high requirements for the steering gear,and the airbag takes a long time to inflate,and the overall size is too large,making it difficult to wear.Aiming at the problems existing in the existing fall detection and protection systems,this paper proposes a fall detection-oriented SVM algorithm based on the split method optimization theory,and on this basis,designs a set of airbag protection systems based on chemical triggering methods to achieve Real-time detection and protection of fall behaviors can improve the recognition accuracy,sensitivity,and specificity of the system.The main research contents of the thesis are as follows:（1）Analyze the fall process of the human body and divide it into four stages:balance,imbalance,weightlessness,and touchdown.Combining the relative mass distribution table of various parts of the human body,the waist is selected as the best wearing position of the sensor.（2）The hardware design of the fall detection system.The fall detection system designed in this paper uses the STM32F103ZET6 microcontroller transplanted with theμC/OS-Ⅱsystem as the core,uses the WT901C sensor to collect the acceleration,angular velocity,and attitude angle data of the human body,and smoothest the data and stores it in the designed Micro-SD after Kalman filtering.In the card,the SIM868 module is used to realize the coordinate of the fall position and the sending of the help message,and it is powered by the 800m Ah rechargeable lithium battery.（3）Dimensionality reduction of parameters reflecting the characteristics of human falls and daily activities.The dimensionality of the 7 initial characteristic parameters is reduced by the split method,and the optimal characteristic parameter is the cumulative change of the total acceleration acc^Cum,the change rate of the total acceleration acc^MAD,and the angular velocity Amplitude and cumulative changegsma^Cum,closing attitude angle changeΔΩ_cc.（4）Algorithm design for fall detection.This paper uses MATLAB tools to analyze the collected simulated falls and daily activity data and designs a fall detection algorithm based on split method and support vector machine.The algorithm uses the data collected by sensors to train the classification model in the initial feature space,and then uses the split method to reduce the dimensions of the features to obtain the classification model on the optimal feature space,and finally the classification model on the optimal feature space on the single-chip microcomputer Perform reconstruction to realize real-time detection of fall behavior.（5）Airbag protection system design.The various components of the protection system are designed and processed into real objects.The working principle of the protection system is that when the algorithm determines that the human body falls,it outputs a 150m A current through the serial port to detonate the electric fuse,and uses the thrust generated by the explosion to push the striker forward.Pierce the carbon cylinder to inflate the airbag to achieve the purpose of fall protection.The overall size of the protection system is 58*14*21mm,and the overall weight is about 93g.The experimental results show that the SVM algorithm based on split method has a good classification effect on falls and daily behaviors,and its recognition accuracy,recall rate and specificity are 97.3%,99%and 96.1%,respectively,which are better than the threshold method of 93.67%,90.83%and 95.56%.Through the fall time test and the airbag inflation time experiment on the human body,the average time of the weightless phase of the human body is 806.93ms,and the average inflation time of the protective airbag is 350.4ms.The airbag can be opened before the human body falls and touches the ground,thus verifying the effectiveness of the protection system.