Context-Aware Sensing and Implicit Ground Truth Collection: Building a Foundation for Event Triggered Surveys on Autonomous Shuttles
The LINC project aims to study interactions between passengers and autonomous vehicles in natural settings at the campus of Technical University of Denmark. To leverage the potential of IoT components in smartphone-based surveying, a system to identify specific spatial, temporal and occupancy contexts relevant for passengers’ experience was proposed as a central data collection strategy in the LINC project. Based on predefined contextual triggers specific questionnaires can be distributed to affected passengers. This work focuses on the data-based discrimination between two fundamental contexts for LINC passengers: be-in and be-out (BIBO) of the vehicle. We present empirical evidence that Bluetooth-low-energy beacons (BLE) have the potential for BIBO independent classification. We compare BLE with other smartphone onboard sensors, such as the global positioning system (GPS) and the accelerometer through: (i) random-forest (RF); (ii) multi-layer perceptron (MLP); and (iii) smartphone native off-the-shelve classifiers. We also perform a sensitivity analysis regarding the impact that faulty BIBO ground-truth has on the performance of the supervised classifiers (i) and (ii). Results show that BLE and GPS could allow reciprocal validation for BIBO passengers’ status. This potential might lift passengers from providing any further validation. We describe the smartphone-sensing platform deployed to gather the dataset used in this work, which involves passengers and autonomous vehicles in a realistic setting.