Along with the proliferation of smartphones, smartphone context-aware applications are gaining more and more attention from manufactures and users. With the capability to infer user's context information i.e., if the user is in a meeting, driving, running or at home, smartphone applications can react accordingly. However, limiting factors such as limited battery capacity, computing power and inaccuracy of inference caused by the in-accurate machine learning models and sensors hinder the large deployment of context-aware applications. In this master thesis, I develop CompleSense, a cooperative sensing framework designed for Android devices that facilitates the establishment and management of cooperation group so that developers can further exploit the potentials of cooperative sensing without worrying about the implementation of system monitoring, data throttling, aggregation and synchronization of data streams and wireless message passing via Wi-Fi. The system adopts Wi-Fi Direct technology for service advertisement and peer discovery. Once the cooperative group is formed, devices can share sensing and computing resources within short range via Wi-Fi connection.
CompleSense allows developers to customize the system based on their own optimization needs, e.g., optimizing the trade-offs of cooperative sensing. System components are loosely coupled to ensure extensibility, resilience and scalability of the system, so that failure or change of a single component will not affect the remaining parts of the system. Developers can extend from the current system by adding customized data processing kernels, machine learning models and optimized sharing schemes. In addition to that, CompleSense abstracts the controlling logic of sensors, developers can easily integrate new sensors into the system by following a pre-defined a programming interface.
The performance of CompleSense is evaluated by carrying out a cooperative audio similarity calculation task with varied number of clients which also confirms that CompleSense is feasible to be deployed for lower tier devices, such as Motorola Moto G.
