Research On Fall Detection Algorithm For The Elderly Based On Muti-sensor Fusion

With the development of aging society,and the increase in empty-nest families,accidental falls to the elderly also increased.Timely assistance to the elderly fell,will effectively reduce the disability and mortality caused by falls.Therefore,the fall detection system and algorithm for the elderly has become a hot issue to study through science and technology to detect falls,as much as possible to reduce the elderly damage.Wearable fall detection systems have the following characteristic: users can bring it indoors and outdoors easily,which can satisfy the demand of fall detection system for elderly.Based on the wearable system,the fall detection algorithm is studied in this paper.The main research is as follows:1.For high rate of false positives of fall detection algorithms,which based on acceleration and attitude angle threshold,this paper combines plantar pressure with acceleration and attitude angle information,which inspired by the gait analysis in bio-medical field,and puts forward a Fall Detection algorithm based on Multi-Behavior fusion(MBFD).In MBFD,if acceleration and attitude angle exceeds than the thresholds,carry on plantar pressure threshold determination;then two pressure matrices are constructed and calculated.If both matrices are equal to zero,it is judged as fall.The rest cases are regarded as non-fall.The optimal thresholds of acceleration,attitude angle and plantar pressure in the algorithm are determined by PSO algorithm.Simulation experiments proved that MBFD has a low false positive rate of bend and squat,and has a higher correct rate of fall recognition.2.The threshold-based fall detection algorithms have more dependence and the effect of the individual behavior difference on the thresholds.To solve this problem,a new Fall Detection Algorithm based on Support vector machine(SFDA)is proposed.SFDA first performs processing on the original behavior data,which is transformed into 18-dimension behavioral feature vectors.then the optimal parameters are selected by k-fold cross-validation.The behavior feature vectors are trained through the optimal parameters.The behavioral test set is predicted by the training model to obtain the behavioral feature label to judge the fall behavior.The simulation results show that the fall behavior is effectively detected by SFDA,and the reliability of the algorithm is verified.3.For better fall detection algorithm described above,the fall detection system based on cloud server is designed.The system consists of wearable device and cloud server part.Foot,chest and GPRS module of wearable device upload the collected behavioral data to the cloud.In cloud server,LNMP is the basic operating environment,and the uploaded data is processed by decoding,storage and detection and training of SFDA,then display on Web.The experiment shows that the feasibility of SFDA in the fall detection system based on cloud server,and the system has a high accuracy for certain specific behaviors.