Research On The Daily Behavior Monitoring System For Old Man Based On Probabilistic Reasoning

Cognitive decline will seriously affect the quality of life of the old man and impede the sustainable development of health care.Cognitive neuroscience experts have demonstrated that some abnormal behaviors are the significant indicators of cognitive decline.Therefore,monitoring the daily behaviors of the old man can help early diagnosis of cognitive impairment symptoms and improve the health of old man.This dissertation mainly researches the elder's daily activity recognition and abnormal behavior detection.Its tasks are as follows:(1)According to the existing problem that probabilistic statistics or ontology methods cannot express complex structural relationships and uncertain knowledge at the same time,this dissertation proposes a method for recognizing the daily activities of the elder based on log-linear probabilistic ontology.The daily activities,according to its complexity,are divided into multiple levels.We combine the description logic and log-linear model to construct OWL2 probabilistic ontology,which is analyzed and enhanced to infer the daily activities of each level in real time by using the java framework.Compared with non-probabilistic ontology methods,the proposed probabilistic ontology approach combines symbolic logic and probabilistic reasoning closely.It can use contextual information to infer daily activities which are currently most likely to be performed without sacrificing the advantages of ontology modeling and probabilistic reasoning.The recognition framework supports the representation of heterogeneous and inconclusive contextual information.The experimental results show that this method has greatly improved the accuracy of recognizing daily activities and acquired relatively high accuracy and recall.(2)Most of the abnormal behavior detection methods can not provide detailed semantic description of abnormal behavior,and also lack of consideration of personal habits and characteristics of the elder.Therefore,this dissertation proposes a new hybrid abnormal behavior detection method which includes short-term abnormal behavior detection and historical behavior analysis based on logic and statistical techniques.Firstly,the Markov logic network inference engine is detected to the start and end time of the activity from the series of consecutive multiple activity events.Secondly,the boundaries of activities and the recognized daily activities are regardedas the input information by the knowledge inference engine to output the short-term abnormal behavior.Thirdly,the frequent pattern mining technology is used to mine and analyze the behavior patterns of the elder in the past period time to output long-term abnormal behavior.Compared with statistical-driven or knowledge-driven methods,the hybrid logic and statistics-based abnormal behavior detection method proposed in this dissertation can effectively represent the ambiguity of domain knowledge and process activity boundaries to acquire detailed semantic description and long-term trend of abnormal behavior.The experimental results show that short-term abnormal behavior can be detected more accurately when the boundaries of activities are correctly recognized.The specific long-term abnormal behavior cycle that deviates from the regular pattern of daily activities can also be distinguished.