Research On Elderly Posture Recognition Algorithm Based On BP Neural Network

Population aging is a global trend. More and more elderly people need to be taken care of, while the number of nurses is much smaller than the growing number of elderly. The elderly care has aroused the concern of many experts and scholars. Posture recognition as an important direction for elderly care has important theoretical significance and practical value.In this paper, the research status of the elderly posture recognition is introduced. And we analyze the technology based on posture recognition, common posture recognition algorithm and standards for algorithm performance. As conventional posture recognition system uses a single sensor or a plurality of single type sensor, the type of posture recognition is less and recognition accuracy is not high. We propose a posture recognition system which use heterogeneous sensors for elderly posture recognition.As in actual experiments it is found that the thresholds of multi-threshold classifier are difficult to select and the classifier has less adaptability to different volunteers and low the correct rate. In the data processing system, we give detailed feature extraction, normalization, data diversity and other data preprocessing steps. Then we propose posture recognition algorithm based on BP neural network, effectively avoiding the problem existing in a multi-threshold posture recognition algorithm.As part of the posture recognition system’s hardware nodes placed on older people for sending data from the sensors have a relatively large volume and are inconvenient to carry, affecting the normal activities of the elderly, we design a elderly posture recognition system solutions based on wireless sensor networks with the miniaturized nodes. The system structure, hardware design and software process are given in detail to build a posture recognition algorithm implementation platform.Based on the above platform, we collected 10 volunteers a total of nearly 30,000 posture data to test multi-thresholds posture recognition algorithm and the posture recognition algorithm based on BP neural network. The test results show that the two kinds of posture recognition algorithms both can identify four kinds posture:the standing, sitting, walking and falling down. However, the average correct rate of multi-threshold posture recognition algorithm is only 39.6%. The proposed algorithm in this paper has a correct average rate of 98.5%.