Based On Single Node Inertial Sensor Human Behavior Monitoring

Due to changes in ways of working and the improvement of living standards, people’s living habits and eating habits are changing.On the one hand, people stay inside for a long time, travel basic rely on transport, lead to a lot of people are lack of exercise, and there is a long wait for a phenomenon.By monitoring the body’s daily movement, reasonable motion planning has important application value, on the other hand, with the proportion of older people increasing each year, the empty nest elderly health problem has become the focus of social concern, in the old man living alone or without accompanying family members, accidentally fell and the old man’s closely linked to the high fatality rate.Through the fall detection can reduce falls happen with health care to the time interval between the serious effect of reducing long-term "lying".Research of convenient, economic and accurate of the old man fell and automatic detection device has great practical research value.Due to adopt the method of multi-sensor monitoring human behavior have to wear uncomfortable, large amount of calculation, the problem of high cost, human behavior recognition based on single node inertial sensor has higher application value.But now is not a unified standard regulations specific behavior of sensor location, wearing different research will wear in different parts of the sensor nodes.To solve above problems, this paper designed the experiment of inertial sensor based on single node each part of the human behavior recognition performance comparison, finally it is concluded that the fall in the detection, walk, run, up the stairs, down the stairs, static posture, etc, the waist is the most ideal inertial sensor worn parts, finally implemented on embedded devices and improve the human behavior under the waist deployment location monitoring algorithm.The main content of the paper work for:(1) production based on inertial sensor monitoring nodes of human behavior.The node by the sensing module, a microprocessor, a storage module, communication module, real time collection, storage, sending, the inertial sensor data when dealing with human behavior.As envoys point can real-time monitor the human behavior, improve the monitoring nodes range, reduce the power consumption of the monitoring node is also the work of this paper.(2) the experiment is human body each node of the gait, fall, static posture recognition performance.Experimental design data collection and acquisition section different people different body parts of human behavior, and then based on the collected data on the Matlab design behavior recognition algorithm, including the gait recognition, fall detection, static posture, walk for recognition.Algorithm design mainly includes the following contents: the design of filter, feature extraction and selection, the design of classifier.Choose butterworth low-pass filter to filter sensor data.Choose three axis acceleration of FFT as gait feature;Selects the impact, gesture, body flip Angle after the fall, the speed of the vertical downward, weightlessness features as a fall;Activity and body posture as static gesture recognition characteristics, choose to use the peak valley difference and the difference of time as recognition walked several characteristics.Joint use of decision tree and support vector machine(SVM) to identify the actions of human body.Finally concluded that the waist is identifying the gait, fall at the same time, the number of static posture, walk the most ideal position.(3) in human behavior under the way of monitoring nodes to achieve the waist deployment behavior recognition algorithm.Will design the behavior of the monitoring algorithm is transplanted to embedded devices, and to increase the recognition rate of the stairs, down the stairs, the height of the barometer is used to measure the body change value, will change the height of the human value as one of the gait features.Finally design experiments to verify equipment performance and the algorithm performance, the results show that gait recognition rate is higher than 99%, sensitivity and specificity of 99.50%, falls steps meter accuracy is better than that of the existing meter products.