Human Bodies' Fall Prediction And Direction Recognition Based On HMM-SVM

With the aggravation of the world population aging and the dramatic increase in empty-nesters,the elderly,who suffer from physical function decline and decrease of reactivity,are apt to experience falls.The fall prediction system not only informs relatives and the monitor center immediately,which reduces the duration when the elderly wait for rescuers,but also provides sufficient time to boot airbags before human bodies impact the ground.Therefore,the system minimizes damage human bodies suffer from falls.It is conducive to reducing medical expenses and enhancing the elderly's confidence of independent living.Most research only detect the fall trend before the impact,but they are unable to recognize the fall direction further.The identification of the fall direction facilitates opening the specified airbag to provide targeted protection.In this paper,the problem of multi classification for the fall direction was studied.A multi classification algorithm,with sufficient lead time and high detection rate,was proposed.This is the key to fall protection system,and is also the focus of this study.An inflatable device based on the air nozzle and the steering gear was also constructed.Compared to the traditional device,it is more lightweight.The above two points are the innovation of this study.Firstly,the system composition was determined.The MEMS inertial sensor,worn at the waist,was used to acquire data and the smartphone was responsible for implementing the algorithm.By analyzing the acceleration and angle's time domain characteristic curves of falls and activities of daily living(ADLS),it proved the feasibility of the fall prediction and direction recognition before the impact;the wrapper method was employed to select the feature,and the optimal feature combination was determined.Secondly,The fall predict algorithm based on hidden markov model-support vector machine(HMM-SVM)hybrid model was realized.Probabilistic models,which represent 4 types falls,were established by Matlab's HMM toolbox to generate matching feature vectors.The parameter optimization and the discriminant model,based on SVM for matching feature vectors to predict falls and recognize directions,were completed by the LIBSVM development kit.Finally,the fall predict application was developed in Eclipse environment and the inflatable device was designed.Then,the HMM-SVM multi classification algorithm and the whole machine were tested.The experiment results show that 4 types falls can be predicted at least 245ms ahead the occurrence of the impact with the accuracy of 96%,98%,96%and 92%,respectively.The recognition rate of ADLs is 97.5%.The approached algorithm's high detection rate and adequate warning time are verified.The test result of the whole machine confirms that the air bag has been filled before the impact.It proves the rationality of the structural design and the feasibility of the system.