The Transition Cube Matrix: Due To The Changing Of Population Structure

The number of elder people is increasing due to the changing of population structure. The active people working outside could not take care their elderly all the time. In most cases the elder people have to live alone in their home. The technologies for assisting older people able to live alone at home is become necessary. Several studies [][] have focused on the monitoring and notification when events causing injuries, such as falling. Other interesting works [] [] introduced the technologies for checking the abnormalities under the concept that if the pattern of activities in daily life are changing it might mean unhealthy, such as not enough sleeping hours, or repeating harmful actions in a pattern. In our work we focus on discovering features
The vision based technique for classifying the 12 basic activities previously proposed will be adopted as the input of model. The concept is that the activities, duration, order of activities, and frequency transition of activities will be observed and modeled into a comprehensive matrix, so called, the TCM. However, in real daily life a pattern of activity cannot be predicted. Our work doing based on the concept that pattern of daily life activity not same in a day but similar activity in similar times. Under this concept, we defined daily activities as normal activity. The normal activity was compared with itself but change some activity different that we call noise. In our method, we choose the Root Mean Square Error (RMSE) for compared the difference of these 2 data. The RMSE has been used for judging the performance prediction model. We applied this measurement quality by applying the prediction value as in consideration activity and the actual value as based on normal activity. Sometimes times in the activity is a discrepancy. We use sliding windows for estimate the difference between same and similar times to use the result obtained for analyzed normal or abnormal activity in similar times. We testing this method at the size of windows = 5. A window in sliding windows collects 1 matrix cube every 1 hour
Rashidi and D. J. Cook [10] proposed mining method for discovering patterns of activities. This method can discover both discontinuous activity and varied-order patterns and they used hierarchal clustering for pattern recognition. The Hidden Markov model (HMM) was applied to activity recognition. D. Lymberopoulos et al. [12] introduced model in home sensor network for Extracting spatiotemporal human activity patterns by using an infrared sensor for detecting person's movement over space. The pattern can discover from sequence of time and duration when elder transition from room to room.
The pattern of activities is one part in the abnormal activities detection in of daily living. H. Jung et al. [13] developed a new abnormal detection method in a life pattern by using an alignment method when changing sequence of activities in daily life.
The support vector data description (SVDD) was applied for abnormal pattern detection. J. Shin et al. [14] use this method for automated abnormal behavior analysis by using information from infrared motion sensors. M. Novak et al. [20] apply Neural network Self-Organizing Maps (SOM) for anomaly detection following in three types of anomalies in human behavior. The daily pattern of user was learned and detected abnormal