Research And Application Of Human Fall Detection And Early Warning Algorithm Based On Android Mobile Phone

In the increasingly serious population aging population,falls threat old man's physical and mental health seriously.It's no doubt that the fall causes secondary damage for the empty nest elderly when they fall,but don't receive timely and effective rescue.Therefore,it is of great significance to improve the quality of life of the elderly by studying the fall detection,warning algorithm of dangerous actions and developing the related health support products.However,the detection rate of the existing detection method had low and limited to the fall detection,lacking dangerous action warning,meanwhile,the detection system on the hardware circuit requirements are high,so that the majority of users is difficult to accept.But the fall detection based on the Android mobile phone has high detection accuracy,good portability and the advantage of real-time detection with the prospect of practical application and great research value.In this paper,we studied the two-class union detection algorithm using Android mobile phone as the carrier,and designed the system application for fall detection and early warning to alleviate the social problem of the elderly fall.The main works of the paper are listed as follows:(1)The 3D model of human body movement is established,and the human body abstraction is simplified as a single rigid body based on hip movement.And the paper gave a deep analysis on the variation of acceleration,angular velocity,angle and energy in the process of falling.(2)The human posture experiment is analyzed and designed carefully,a data acquisition system based on Android built-in sensor is developed,which achieved human motion waveform cross screen display,data storage rapidly,parameter settings and 3D Box demonstration.This paper focuses on the completion of the experimental data using the median filtering pre-treatment and normalization.Meanwhile,the characteristic values of fall detection and motion warning are selected,including the maximum acceleration,the maximum angular velocity,angular variation and acceleration amplitude area.(3)Combining with the statistical machine learning strategy,a two-level joint detection algorithm based on Android mobile phone is proposed.First-level fall detection algorithm based on SVM namely support vector machine realized the classification of fall and non-fall action.The second-level early warning algorithm based on the GBDT namely gradient boosting decision tree realized the specific classification of the warning action,including the excessive lean forward,excessive lean backward,excessive lean leftward,and excessive lean rightward.The effectiveness of the detection algorithm is evaluated by three indexes: the correct rate,false positive rate and false negative rate.The results show that the three indexes of the fall detection algorithm are 92%,7% and 1% respectively,the average of the early warning algorithm are 92.5%,7% and 0.5% respectively.(4)Two kinds of health support products related fall were independently developed,using the detection algorithm.Development kit is the anti-fall warning system based on the Android phone and the fall alarm based on MTK platform,while the system realizes multi-scene,multi-repeation precise positioning,track display and local,remote alarm and dangerous alarm early warning and other functions by the elderly anti-fall warning remote monitoring platform with the help of the micro-cost investment.In addition,the test results of system software function performed as expected,and achieved good user experience,which meet the system stability and fluency requirements.The result of running algorithm on the system shows that the fall detection algorithm predicts the correct rate of 90%,the average accuracy of early warning algorithm is 91.25%.