Adaptive Confidence Evaluation Scheme for Periodic Activity Recognition in Smart Home Environments
Yi-En Tan, Chun‐Chih Lo, Chin‐Shiuh Shieh, Denis Miu, Mong‐Fong Horng · 2019
The concept of Internet of Things (IoT) and related applications have developed rapidly worldwide in recent years. Application such as smart home has made it possible to monitor, control and notify alerts in a home. It also able to improve an individual's quality of life by discovering periodic human activity to enable home automations. Furthermore, the selection of appropriate set of sensors and correct preprocessing of sensor data are all important elements to correctly identify user's daily activities. However, these application usually ignore the spatial information that each sensor may provide, and identify user activities simply based on the data collected from sensors. Moreover, the confidence which indicates the probability that activities are correctly recognized is also an important issue. To enhance the result of periodic activity recognition, only the results with high confidence should be included when training sensor data. Thus, this paper proposes an adaptive confidence evaluation scheme that evaluate activities dynamically to enhance the result of periodic activity recognition. Experiment results showed the proposed method with adaptive confidence evaluation scheme may reduce the number of discovered activities at an average of 17.87% and improve the accuracy of find the occurrence of activities compared with proposed method without adaptive confidence evaluation scheme. This indicate that the proposed method with adaptive confidence evaluation scheme can effectively applied to smart home environments.