UC Irvine

The ubiquitous IoT devices have promoted the growth of applications that require target positioning. While precise positioning remains a challenge for IoT devices, existing methods mainly focus on optimizing the positioning algorithms or expanding the sensing modalities. One opportunity area of improvement is to leverage their wireless communication capability to exchange sensing data of different modalities with nearby devices. This work proposes collaborative sensing frameworks to enhance the accuracy of position inference. Our approach to collaborative sensing is divided into selection of reliable devices, the data-exchange mechanism, and the data-fusion algorithm. We demonstrate our approach on two different position-sensing systems: one for chest compression-depth estimation and one for indoor pedestrian localization. The key components used in the frameworks include quality assessment of sensor data using mutual information, data-fusion algorithms based on the chest-compression model and the pedestrian-encounter model, and the data-exchange mechanism using Bluetooth Low Energy (BLE). A real-time position-estimation method based on our chest-compression model is proposed to remove the noise and handle the cumulative error. A collaborative conditional random field algorithm is developed to reduce the convergence distance in the localization estimation. Experimental results show our collaborative-sensing approach achieves higher rate of convergence than the existing solutions in terms of real-time estimated position. This dissertation proposes an alternative approach to optimizing the position-sensing systems through collaborative sensing. Our work represents the first step towards the universal positioning on IoT devices and provides research opportunities for further improvement and exploration.