International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, Volume XXXIX-B4, 2012
XXII ISPRS Congress, 25 August — 01 September 2012, Melbourne, Australia
THE DIGITAL CAMERA AS A GPS COMPASS:
LOCATION MATCHING USING POINT CLOUDS
Masafumi Nakagawa*, Daisuke Kato, Hirotaka Endo, Hiroshi Itaya, Kenta Ochiai, Anna Nakanishi
Shibaura Institute of Technology, 3-7-5, Toyosu, Koto-ku, Tokyo, 135-8548, Japan - mnaka@shibaura-it.ac.jp
Commission IV, WG IV/2
KEY WORDS: Seamless positioning, Point-cloud, Image-based GIS, Panorama image, Image matching, Camera calibration
ABSTRACT:
We focus on the potential of a camera to act as a location sensor, assisting other location sensors to improve positioning accuracy. A
camera is installed in almost all mobile devices. Moreover, the camera can be used as a location sensor without additional
transmitters or receivers. However, if the camera is used as a location sensor, reliable maps will be required. Although there are
some location-matching approaches that use maps, the success rate of location detection depends on the representation of a 3-D
model and its information content. Compared with a model representation based on Computer-Aided Design (CAD), a point-cloud
representation is more photorealistic. We therefore focus on point-cloud data being used for reliable maps. Our proposed location-
matching methodology is based on image matching using images from a digital camera and panoramic images generated from a
massive point-cloud in an image-based Geographic Information System (GIS). We conducted experiments in location matching
using a digital camera to supply the input data for location detection and a point cloud taken from a terrestrial laser scanner. We
have confirmed that our approach can detect locations using a digital camera that is restricted to horizontal movement.
1. INTRODUCTION
Recently, 3-D data-acquisition techniques and seamless
positioning techniques in indoor and outdoor environments
have become necessary for the construction and maintenance of
facilities such as roads, bridges, tunnels, and public buildings.
Laser scanning is one of these 3-D data-acquisition techniques.
A laser scanner can acquire a 3-D point cloud by measuring the
distance to a surface for a range of scanning angles. Moreover,
a calibrated digital camera with a laser scanner can be used to
acquire color information to add to the point-cloud data.
Massive point-cloud data taken from terrestrial laser scanners
are used for various visualization and 3-D modeling
applications such as GIS data acquisition in urban areas and
environmental investigations.
For many fields such as navigation, disaster relief, and
construction automation, the Global Positioning System (GPS)
is used to identify sensor locations. Although the GPS is a
convenient system, it is restricted to outdoor environments.
However, a seamless positioning technique for both indoor and
outdoor environments is required to obtain sensor locations.
Currently, mobile devices have many location sensors such as
Radio Frequency Identification (RFID) tags, Bluetooth, and
wireless LAN [1]. Practical issues with the above systems for
facility monitoring are accuracy requirements and hardware
installations around indoor-outdoor environments in urban areas.
Magnetic direction sensors can acquire directions directly.
However, the magnetic field is affected by metallic materials
such as iron frames in indoor environments. In addition,
although the simultaneous localization and mapping technique
[2, 3], based on laser scanning, can achieve accurate sensor
localization, laser scanners are too large to mount on mobile
devices in their current state of development. Therefore, we
focus on the potential of a camera to act as a location sensor,
assisting other location sensors to improve positioning accuracy.
A camera is installed in almost all mobile devices. Moreover,
the camera can be used without the need for transmitters or
receivers.
However, for a camera to be used as a location sensor, reliable
maps will be required. Although there are some location-
matching approaches that use maps such as 3-D Computer-
Aided Design (CAD) models and image data sets, the success
rate of location detection depends on the representation of 3-D
model and its information content. In addition, estimation of the
location as the external orientation parameters of the camera
requires the matching of corresponding points in camera images
and the 3-D model. Corresponding-point detection is easy in
manual processing. However, corresponding-point detection is
not simple in fully automated approaches, because the features
are inherently different (e.g. edges in images and boundaries in
3-D models).
Compared with a model representation based on a Triangulated
Irregular Network (TIN) or on CAD, a point-cloud
representation is more photorealistic. Moreover, the rendered
result of a massive point cloud can be used as a panoramic
image. We therefore focus on massive point-cloud data for use
in reliable maps. Our proposed location-matching methodology
is based on image matching using an image taken from a
camera and panoramic images generated from a massive point
cloud in an image-based GIS.
Our development aims for 10 cm accuracy to assist with
existing indoor positioning techniques. When facility
information for construction and maintenance is geocoded onto
maps, higher accuracy and higher spatial resolution are required.
We therefore describe fine location matching in this paper. We
then develop this matching approach to confirm that a camera
can be used as a location sensor in a fully automated procedure.
Finally, we present experimental results that confirm the
validity of our approach.
