The Research And Implementation Of Intelligent Human Fall Detection System Based On Multi-sensor Collaboration

With the growth of age,the elderly are most likely to fall because the physical condition of the elderly,the ability to sense and recognize the outside world becomes weaker and weaker.Falls are rarely fatal,but can cause non-fatal injuries such as broken bones,bruises,sprains,and dislocated joints.If the elderly do not get the first medical attention,it can worsen the situation further,from non-fatal injuries to fatal injuries.Therefore,it is of great practical significance to design a fall detection system,which can send out remote alarm information when human body falls.In this thesis,an intelligent human fall detection system based on multi-sensor collaboration is researched and implemented.Main functions of the system include two aspects.On the one hand,it can automatically identify the position of the fall detection device.On the other hand,it can select suitable feature data to detect falls according to different positions.The system mainly uses accelerometer,gyroscope and magnetometer for data collection,and cooperates with three kinds of sensor data for wearable position recognition and fall detection.Main research contents is as follows:Firstly,Selection of sensor data features.We use the method of moving average filter to remove the noise of sensor data.Then,the change of acceleration,angular velocity and attitude angle is analyzed under different behaviors and different wearing positions,and the characteristics of combined acceleration,combined angular speed and posture angle are constructed.Eight kinds of time domain and frequency domain information such as mean,variance and maximum value are extracted to describe the above three characteristics.Feature data sets are used to train algorithm models and identify system classification.Secondly,a wearable position recognition algorithm of fall detection equipment is designed and implemented.According to the characteristics of different wearing positions,a multi-classification based on Softmax regression algorithm is proposed to identify the wearing positions,and the UMAFall data set was used to train and verify the algorithm.Thirdly,fall detection algorithm is designed and implemented.According to the characteristics of different wearing positions,there are differences between daily behaviors and falls.The Filter feature selection method is used to select the characteristics suitable for different wearing positions.Then KNN algorithm is used for detecting fall,and the UMAFall data set is used to train and verify the algorithm.Finally,the hardware system platform is designed,which includes data acquisition module,processor module,power module,GPRS/GPS module and alarm module.Moreover,the algorithm performance of the hardware system platform is tested.It founds that the wearable position recognition algorithm and the fall detection algorithm of the fall detection device have high accuracy.