Rise of the Indoor Crowd: Reconstruction of Building Interior View via Mobile Crowdsourcing

Crowdsourcing is a technology with the potential to revolutionize large-scale data gathering in an extremely cost-effective manner. It provides an unprecedented means of collecting data from the physical world, particularly through the use of modern smartphones, which are equipped with high-resolution cameras and various micro-electrical sensors. In this paper, we address the critical task of reconstructing the indoor interior view of a building from crowdsourced data. We propose, design, and prototype IndoorCrowd2D, a smartphone-empowered crowdsourcing system for indoor scene reconstruction. We first formulate the problem via trackable models and then employ a divide and conquer approach to address the inherently incomplete, opportunistic, and noisy crowdsourced data. By utilizing the image information and sensory data in a coordinated way, our system demonstrates high result-accuracy, as well as allows a gradual build-up procedure of the hallway skeleton. Our evaluation result shows that IndoorCrowd2D achieves a precision around 85%, a 100% recall and a F-score around 95% for reconstructing college buildings from 1,151 datasets uploaded by 25 users. This reveals that our image and sensor hybrid method is more robust to overcome errors and outliers as compared to image-only method.