Mobile crowdsourcing is a potential enabling technology to revolutionize large-scale pervasive sensing in an extremely cost-effective manner. It provides an unprecedented way of collecting data from the physical world, particularly through the use of modern smartphones which equipped with high-resolution cameras and various micro-electrical sensors. In this work, we address a critical task of reconstruct indoor building interior view and hallway skeleton from crowd-sensed data. We propose, design, and implement IndoorCrowd2D, a smartphone empowered cloud-based crowdsourcing system for large-scale indoor scene reconstruction. IndoorCrowd2D fills a gap in current indoor scene reconstruction systems by leveraging crowdsourcing to collect large-scale data. To overcome data quality challenges posed by untrained individual contributors, we propose an efficient scheme to enhance the captured data quality for building interior reconstruction at mobile front-end. At the cloud side, we deploy an automated image vector bundle (IVB) processing pipeline, which generates building interior views from images and sensory data. Moreover, we provide an estimated building skeleton for each building floor which shows the relative position of each panorama image.
IndoorCrowd2D is comprised of two components: (i) a mobile phone app (prototyped for Android) and (ii) a cloud computing backend, which is deployed on an infrastructure- as-a-service (IaaS) cloud. The app allows users to shoot and upload building interior scenes annotated with synchronized sensor data including last known GPS position and compass, gyroscope, and accelerometer readings. The cloud service receives many crowdsourced datasets simulta- neously, and for each set it leverages computer vision and sensing algorithms to continuously model the interior and generate an image vector bundle. Across multiple data sets, IndoorCrowd2D utilizes a building interior and floor map skeleton reconstruction pipeline, and ultimately provides a visually appealing interior view of each building.
As a crowdsourcing system for developing interior views and skeleton reconstruction, IndoorCrowd2D has the potential to help existing online map providers extend their service to indoor environments at a large scale. By providing building geographical information to the server, for example, IndoorCrowd2D allows the user to download the existing indoor panoramic view and geo-data for each individual level inside the building. Users may also port our data to existing online maps (such as Google Map) by use of their APIs. Once fully hardened, IndoorCrowd2D can be deployed as a service providing indoor panoramic and geo-data for each individual level of buildings from all over the world. Our mobile application can be viewed as a useful, portable tool that encourages people to collect and contribute indoor-spatial data to the cloud. More generally, we believe that IndoorCrowd2D will extend existing online map service to indoor environment at a large scale. With the rapid development of virtual reality (VR) devices, in the future, one might utilize such device to connect to our system and take a virtual tour of a building almost anywhere in the world.
IndoorCrowd2D serves as an important stepping stone towards the ultimate goal of economically-viable massive indoor 3D model reconstruction.
For more details, please check our SenSys 2015, ICDCS 2015 paper and Infocom’14,15 Poster
Rise of the Indoor Crowd: Reconstruction of Building Interior View via Mobile Crowdsourcing,
Si Chen, Muyuan Li, Kui Ren, Xinwen Fu, Chunming Qiao, in Proceedings of The 13th ACM Conference on Embedded Networked Sensor Systems (SenSys 2015), 2015
[pdf | slides | cite]
CrowdMap: Accurate Reconstruction of Indoor Floor Plans from Crowdsourced Sensor-Rich Videos,
Si Chen, Muyuan Li, Kui Ren, Chunming Qiao, in Proceedings of The 35th IEEE International Conference on Distributed Computing Systems (ICDCS), 2015
[pdf | slides | cite]
IndoorCrowd2D: Building Interior View Reconstruction via Mobile Crowdsourcing,
Si Chen, Muyuan Li, Zhan Qin, Kui Ren, in Proceedings of the IEEE Conference on Computer Communications Workshops (INFOCOM WKSHPS), 2015. [link]