Font Size: a A A

Research On Human Fall Detection Algorithm Based On Machine Learning

Posted on:2022-02-18Degree:MasterType:Thesis
Country:ChinaCandidate:X RenFull Text:PDF
GTID:2518306602992929Subject:Computer Science and Technology
Abstract/Summary:PDF Full Text Request
In recent years,as the trend of population aging in our country has become more pronounced and the number of elderly people living alone has continued to increase,the physical and mental health of the elderly has become one of the common concerns of the country and society.Relevant studies have shown that the high incidence of falls in the elderly and the serious consequences are one of the important reasons for the disability and death of the elderly.Therefore,with the help of modern computer technology,it is very important to be able to detect the fall of the elderly in time and provide treatment.Nowadays,most of the more mature fall detection systems are based on images or wearable sensors,and there is less research on the fall detection of smart phones.Obviously,as a necessity for everyone to go out,smart phones are not easy to forget,easy to carry,and easier to monitor the activities of the elderly in real time.Based on the three-axis acceleration information provided by smart phone sensors,this paper analyzes the changes of various sensor data in people’s daily activities,and further studies the related algorithms of fall detection.Since it is difficult to obtain a large amount of fall data for the elderly through experimental simulations,this article is based on the open source UCI data set to complete all the experiments.The main tasks completed in this article are as follows:(1)Since the data collected by the acceleration sensor will be affected by the noise signal generated by the external environment,which will have an impact on the experimental results,this paper uses Kalman filter to filter the data to restore the original change curve of the signal.(2)Through the analysis of the sensor data when the human body falls in the data set,it is found that every time a person falls,it will be clearly reflected in the change of the data curve: after a steady and sharp rise,it will remain stable again.Based on the characteristics of this change,this article uses a sliding window method to discretize all time series,and extract relevant features in each window.Because some features are not obvious in the time domain,in order to better characterize the curve characteristics,this paper combines time domain and frequency domain analysis,selects a group of features to form a feature set,and combines the forward and backward selection method to get the best feature combination.(3)Considering that the mobile phone-based fall detection algorithm needs to be instantaneous,it is necessary to choose an algorithm with low time complexity and high accuracy.In this paper,after several experiments on the UCI data set to compare the performance of each classifier,a personalized combined classifier is proposed.The combined classifier can make up for the performance defects of a single classifier while ensuring the accuracy of classification.In addition,due to various aspects of human height and body shape,when the trained model is applied to a smart phone,the accuracy rate will be significantly reduced.Therefore,the personalized fall detection algorithm proposed in this paper uses the user’s personal activity information as positive feedback information to participate in the model training,and adjusts the combined classifier model after the original data training.Finally,the experimental results show that the personalized combined classifier model has greatly improved the accuracy and recall rate in fall detection.(4)In previous studies,scholars only paid attention to the correct rate of classification,but in practical applications,models that combine the correct rate and recall rate with better performance have higher use value.In the experiment of this article,the fall behavior and daily behavior are analyzed separately,and the experimental data is combined with the correct rate and the recall rate to judge the performance of the model to obtain a more useful classification model.(5)The personalized combined classifier model proposed in this paper is applied to Android mobile devices,and has the functions of issuing warnings and sending text messages to fixed guardians when a fall is detected.
Keywords/Search Tags:Fall Detection, Sliding Window, Personalized Combination Classifier, Machine Learning
PDF Full Text Request
Related items