Leveraging Fog Analytics for Context-Aware Sensing in Cooperative Wireless Sensor Networks

In this article, we present a fog computing technique for real-time activity recognition and localization on-board wearable Internet of Things(IoT) devices. Our technique makes joint use of two light-weight analytic methods—Iterative Edge Mining(IEM) and Cooperative Activity Sequence-based Map Matching(CASMM). IEM is a decision-tree classifier that uses acceleration data to estimate the activity state. The sequence of activities generated by IEM is analyzed by the CASMM method for identifying the location. The CASMM method uses cooperation between devices to improve accuracy of classification and then performs map matching to identify the location. We evaluate the performance of our approach for activity recognition and localization of animals. The evaluation is performed using real-world acceleration data of cows collected during a pilot study at a Dairygold-sponsored farm in Kilworth, Ireland. The analysis shows that our approach can achieve a localization accuracy of up to 99%. In addition, we exploit the location-awareness of devices and present an event-driven communication approach to transmit data from the IoT devices to the cloud. The delay-tolerant communication facilitates context-aware sensing and significantly improves energy profile of the devices. Furthermore, an array-based implementation of IEM is discussed, and resource assessment is performed to verify its suitability for device-based implementation.