Study On Detection And Modeling Of Human Fall

The increasingly severe of the aging of population draw public concern. Mostof the aged people live with themselves, lacking of nursing of their children orguardians. In a ddition, a ged pe ople suffer f rom the low s tamina an d v aguecognition, which may lead to many unpredictable problems, among which, fall isone of the important problems. Fall has been causing huge attentions all over theworld.The r esearch of fall be longs to the study of human m otion, in whichmechanical sensor, pressure sensor, optical sensor, acoustic sensor, inertial sensorand electromagnetic sensor usually be used. Optical sensor and inertial sensor aremost commonly used, so we use this two kinds of sensors to study the fall. In thisthesis, we mainly use optical sensor assisted with inertial sensor.Then, opti cal se nsor, wh ich has th e ad vantages in term s of describeorientations and positions, was used to analyse the morphological characteristicswhen one was falling. Combin ed the har dware conditions of the laboratory anddetailed analysis of this two categories, we chose the marker-based human motioncapture as our basic method. To overcome the disadvantages of previous algorithm,an new algorithm based on Hough Transform was rased to find the markers in thevideo out of focus. W e co mpared t he results got f rom ou r a lgorithm with th eresults came from the commer cial software ProAnalyst to verify the stability andsuperiority of our algorithm.Inertial se nsor, whose adva ntage is the hi gh pre cision in mea suring t heacceleration and velocity, was used to measure the the acceleration and velocity ofthe torso. After detailed analysis of the inertial data, we figure out the pre-impactlead time and t he angle of posture stability of t hree kinds of fall, which will beused to creat the fall model.Additionally, we found that the human body could be divided into three partswhen f alling dow n. Com bined the data of ine rtial s ensor, we propose d t hethree-parts kinematic model of human fall.