Intelligent Daily Activity Recognition Based On Motion Sensor & Application Development Of Elderly People

With the development of ubiquitous technology, the people's demand for comfortable, safe, healthy living style is also increasing. Activity recognition based on acceleration sensor technology plays an important role in health care, human-computer interaction and action director field. At the same time, the activity recognition is becoming one important part of intelligent home. Physical activity and exercise has become one of the adjunctive treatment of elderly patients in chronic diseases. The appropriate exercise is helpful for elderly people to enhance physical function and slow down aging speed. With the help of activity recognition, we can calculate the amount of activity, analyze their activity patterns, increase or decrease an activity on purpose, which is great significance for the health care of the elderly.There are two mainly methods for activity recognition, one is based on computer vision and another is based on triaxial accelerometer. Methods based on computer vision exist susceptible to interference, small range surveillance, invasion of privacy and other defects,the sensor-based technology is having a strong anti-interference, easy to carry, freedom of access to data, protection of privacy and other advantages.We propose an activity recognition method based on motion sensor and HMM.In the light of the motion type and motion characteristic of elderly people,we extract standard deviation(SD), energy for static activities distinguish in set S, correlation coefficients(Corr_VF), accelerometer magnitude peak(Amp), ratio forward(RAF) and ratio vertical forward(RVF) for similar activities distinguish in set D.After the feature extracting phase,we use improved K-means clustering algorithm to generate the observation set and then we define the recognition model for elderly people. We use the Viterbi algorithm to recognize the activities for elderly people after the parameters are trained by Baum-Welch algorithm.The experimental results shows that our approach is can be applied for daily activity recognition of elderly people,the average accuracy of single-continuous activity is 93.4%, specifically the accuracy of similar walking activities is 93.7% and the accuracy of random-composited activity is 91.1%.During the experiment,we also compare the effects of the different wearing part of accelerometer leads different accuracy of activity recognition: the accuracy of central parts such as waist hip is higher than the limbs such as wrist.We design a system which is used for daily activities caring of elderly people and this system provide graphical information of activities statistic and analysis for relatives and personal doctors of elderly people by counting the activities recognition record, and the system also provides the useful information for elderly people caring in daily lives. The system is great significant to improve the level of health and quality of life for the elderly people.