Combining Context Connectivity and Behavior Association to Develop an Indoor/Outdoor Context Detection Model With Smartphone Multisensor Fusion.

The emergence of seamless mobile navigation systems integrating various Internet of Things (IoT) devices has sparked interest in context awareness enhancement technology. In the concept of advanced adaptive integrated navigation technology, the context comprises two key elements: 1) environment characteristics and 2) carrier behaviors, which are not entirely independent, especially in certain scenarios. Leveraging the abundant sensors in smartphones, a model combining context connectivity and behavior association is developed to detect environment scenes accurately with low-energy consumption across outdoor, semi-outdoor, and indoor spaces. The model comprises three main parts: sensor-based support vector machine (SVM), behavior-aided hidden Markov model (HMM), and classifier combination. The parameters of a behavior-aided HMM are adjusted by behavioral probabilities and a specified exponential moving average method. Four classifier combination techniques, including simple average, environment weighted averaging (EWA), environment and behavior-based weighted averaging (EBWA), and stacking, are used to integrate the environment detecting strengths of multiple smartphone sensors. The proposed model is evaluated on a data set collected from a complex building at Wuhan University and achieves a best environment detection accuracy of 94.22% with stacking ensemble technique. The multisensor model outperforms the other three classifier combination techniques, improving detection accuracy by 6.93% compared to a GNSS-supported model. The proposed model has certain advantages over high-recognition accuracy, low-model consumption compared to the main existing environment detection models.