| Falls are the premier cause of injury-related death for people aged 65 and over,and it is estimated that approximately 15%-35% of older people in China experience an accidental fall each year.With the significant acceleration of the ageing of society,there is an urgent need to focus on falls in the elderly population.Traditional falls detection methods are mostly based on contact wearable devices such as three-axis inertial sensors,pressure sensors,gyroscopes and accelerometers,which do not take into account the convenience and comfort of the elderly in their daily lives.In addition,falls detection is limited by the fact that it detects falls when they occur,but once a fall occurs,the damage caused is irreversible.Therefore,non-invasive 3D skeleton data monitoring is considered to analyze the user’s gait characteristics to predict the risk of falls and avoid accidents.The main research components are as follows.(1)A gait feature extraction method based on the 3D skeleton data of the depth camera Kinect was proposed,combining the personalized characteristics of the subject’s individual information,and the spatio-temporal and kinematic parameters of the subject’s gait were extracted analytically to assess and predict their falls risk.(2)The falls risk levels were classified into two categories: high falls risk and low falls risk.Considering the cost of data collection and the fact that high falls risk samples are not easily available,the novelty detection model one-class support vector machine was used to train and evaluate the feature data under an unbalanced data set.Also,the potential of predicting falls risk in the elderly based on Kinect 3D skeleton data was demonstrated.(3)Based on the above research content,a depth camera-based fall risk prediction system was designed and implemented.The system enables the acquisition,analysis and fall risk prediction and early warning of subjects’ gait data by Kinect.The system can be used in the future to provide longterm dynamic gait health monitoring services for the elderly in community and home scenarios. |
