Design Of Human Fall Detection System Based On Acceleration Sensor

The rapid development of technology and social economy, makes people's life have a quality leap. However it also brings negative product, aging population is one of them. It not only leads to a substantial reduction of social labor and labor imbalance, but also takes up more resources to help the old age social and pension. This phenomenon increases the financial burden on society, the government and the family. At present, China is still in the primary stage of socialism, the social situation that old before getting rich makes the burden of aging population become a huge challenge. According to statistics, fall is the first cause of hospitalization and death in non-pathological in 60 years of age or older, and the proportion is still rising. At present, it still lacks of old people's mature products in fall protection, forecasting and warning in the domestic and foreign markets. The misjudgment phenomenon between elderly fall behavior and other similar processes is particularly evident. There are still many key scientific and technical problems to be solved in the field.The human fall detection system based on acceleration sensors includes hardware devices and detection algorithms two parts. In the foundation of hardware device, the paper analyzes fall behavior in both time and frequency domain. The main work of this paper are as follows:(1) From the research background and significance of the fall detection the paper introduces the current socio-demographic status and fall hazard. Combining mainstream research methods at home and abroad, this paper put forward the fall detection system based on acceleration sensor and indicates its portability and real-time.(2) We analyze the changes and features of body posture and acceleration in the process of fall, and choose the regions of human body which characterize the fall. We select STM8 L for the system microcontroller, MPU6050 as sensor data collection,HC-05 for the communications module, use VS2010 to write PC. These build the overall framework of the system. We establish the data set contains daily behavior, fall behavior, and fall-like behavior through experiments.(3) We extract and analyze the features of fall behavior in time domain. On account of the changes in magnitude of combined acceleration and tilt value of gravity direction, we present a fall detection algorithm in time-domain. The experiments prove that the algorithm in distinguishing between human movement and gesture to lie down, has a high recognition effect.(4) In the transform domain, each human behavior information has been analyzed. Using two classical algorithms Fast Fourier Transform(FFT) and Discrete Cosine Transform(DCT) for signal conversion, we respect and compare the result. It extracts the feature which is the result of pre-processed and reducts its dimension in subspace. Based on statistical learning theory and Support Vector Machines(SVM), we use the direct component's and harmonic components' magnitude of the energy value on the optimal classification interface as fall detection threshold, and propose a fall detection algorithm based on frequency-domain analysis.(5) In the sample space, each algorithm has been validated. In this paper, we propose a detection method based on Fast Fourier Transform and 2-D Principal Component Analysis(PCA). It makes up to 99% recognition rate and proves the superiority of the method. Finally, we summarize the shortage of the system and make observations for further work.