A GIS DATA UPDATING METHOD WITH VIDEO

Ruijun Liu, Myriam Servi

eres and Guillaume Moreau

LUNAM, Ecole Centrale de Nantes, CERMA-IRSTV, 1 Rue de la No

e, 44321 Nantes, France

Keywords:

Geographic Information System (GIS), Architecture Facade Reconstruction, User Knowledge, Video.

Abstract:

We present a method for extracting and updating Geographic Information System (GIS) data from video.

This paper uses a ground-view video sequence as input data. In the ﬁrst step, feature points are extracted from

video, which provide the structure information for architecture facade reconstruction; after discarding outliers,

we create the footprint and reconstruct a 3D model. The second step is the back-projection process. In the

third step, we correct the footprint and update it by using user knowledge. Finally, this information is used to

update GIS.

1 INTRODUCTION

Model reconstruction is quite important in Computer

Graphics (CG) research ﬁeld, especially, the recon-

struction of urban 3D model. It can be applied in

Geographic Information System (GIS) ﬁeld. We will

focus on this interdisciplinary research.

A GIS allows geospatial information manipula-

tion as abstraction, acquisition, storage, analysis and

visualization (Den

egre and Salg

e, 2004). It describes

an urban information digitally, as vector data or as

raster data. GIS data quality is an important aspect.

However, with the city’s development, GIS data can

be inconsistent with reality due to building construc-

tion or destruction. Therefore, users have to maintain

data consistency and this need arises from frequent

GIS updating. As stated in (Heipke et al., 2008), up-

dating GIS data should consider two aspects: ﬁrst one

is logical consistency, second one is a comparison be-

tween data and reality.

Many researchs use aerial image to reconstruct 3D

buildings, but there are few studies about the reverse

process, using 3D reconstruction method to update

GIS data (Zebedin et al., 2008). The key motivation

of our research is to focus on a GIS data updating

method only using a ground-based video sequence as

input. This video sequence is easier to be acquired

than aerial images and only requires a handy camera,

which is not a complex equipment. Traditional meth-

ods often use LiDAR to measure building and recon-

struct GIS data (Li et al., 2011).

Our ﬁrst goal is to update GIS data, such as foot-

print. The ﬁrst step of our solution is to capture a gr-

ound-view video, from which, we will extract build-

ing information and remove the noise. The second

step is to use this extracted information to build a 3D

model. The ﬁrst building model is a 3D geolocal-

ized box. Its extent and width are basic, but accurate

enough to update GIS data.

After a brief review of related work in section 2,

we detail our algorithm in section 3 starting with an

overview of the whole pipeline in section 3.1. In sec-

tions 3.2 to 3.4, we present the information extraction

based on data acquisition, the architecture facades re-

construction and the updating process. Preliminary

results are provided in section 4. Conclusions and fu-

ture work are presented in section 5.

2 RELATED WORK

The scientiﬁc issues in relation with our method

mainly include two different types of techniques: ar-

chitecture facade reconstruction and GIS data acqui-

sition and updating.

2.1 Architecture Facade Reconstruction

Existing architecture reconstruction methods can be

divided into three categories:

• rule-based architecture reconstruction methods

uller et al., 2006) and (Kelly and Wonka,

2011),

• image-based interactive architecture reconstruc-

tion methods (Pollefeys et al., 2008),

511

Liu R., Servières M. and Moreau G..

A GIS DATA UPDATING METHOD WITH VIDEO.

DOI: 10.5220/0003856905110514

In Proceedings of the International Conference on Computer Graphics Theory and Applications (GRAPP-2012), pages 511-514

ISBN: 978-989-8565-02-0

 2012 SCITEPRESS (Science and Technology Publications, Lda.)

• vision-based automatic architecture reconstruc-

tion methods (Goesele et al., 2007).

2.2 GIS Data Acquisition and Updating

Until now, images and remote sensing are the main

technologies for GIS data acquisition and update.

(Steinocher and Kressler, 2006) and (Heipke et al.,

2008) provide us two good overviews of existing ap-

proaches. (Heipke et al., 2008) is a summary of the

GIS data updating process from images, focused on

the macroscopical updating process, rather than on

the individual algorithm. Users need to participate in

the updating process such as: creation, deletion, split-

ting and merging of objects, and the modiﬁcation of

GIS data.

3 PROPOSED PROCESS

3.1 Overview

GIS users may encounter circumstances where GIS

data are inconsistent with real data and need to be

updated. In our method we design two scenarios to

complete this task depending on the input data. One

is using only video as an input and the other one is

using video and existing GIS data as inputs. The dif-

ference is the footprint acquisition, but the updating

method is the same. Figure 1 presents the pipeline

of our designed scenarios. First, we reconstruct a 3D

model with the help of the user knowledge by using

the feature points extracted from video, and then we

create the footprint or acquire it from GIS data. Sec-

ondly, 2D footprint and 3D model are back-projected

onto the video. Then, if the footprint is inaccurate,

we interactively correct it by user knowledge. Finally,

updating GIS data is achieved.

At present, the whole process is not real-time. We

have ﬁnished almost all steps of our scenario recon-

structing a 3D model, only the correction part is in

progress. We will provide some details on our future

work in section 5.

3.2 Information Extraction from Video

3.2.1 Feature Points Extraction

In outdoor environments, many elements may oc-

clude the buildings, such as trees or cars, which make

the feature points tracking fail. Therefore, in or-

der to achieve the main algorithm, as a ﬁrst test, we

used a simple mockup. The whole acquisition pro-

cess simulates the outdoor environment. Then we

Figure 1: Pipeline of GIS data update.

use a state-of-the-art real-time Structure from Motion

(SfM) algorithm, Parallel Tracking and Mapping sys-

tem (PTAM) (Klein and Murray, 2007), to obtain a

feature points cloud and camera parameters. PTAM

updates the map correspondences using bundle ad-

justment, and extracts several thousands of feature

points. At the end, we also compute the normal of

each point as a preparation of the reconstruction pro-

cess in this step.

3.2.2 Filtering Process and Partition

Due to, for example, bad video quality or track-

ing algorithm instability, there are some outliers in

the feature points cloud that will cause bad result in

the next step. Therefore, we ﬁlter the noisy points

with RANSAC algorithm (Fischler and Bolles, 1981).

RANSAC is a general parameter estimation algorithm

designed to cope with outliers in the original data. Af-

ter the ﬁltering process, in order to reconstruct each

facade individually, our approach needs to split the

feature points cloud into several parts.

3.3 Reconstruction

Based on the extracted information, our reconstruc-

tion algorithm aims at reconstructing several facades

fast and effectively.

3.3.1 Footprint Generation from Feature Points

3D world coordinate system is deﬁned as follows: the

red line is X-axis, the green line is Y-axis and the

white line is Z-axis. All facades are parallel with the

Z-axis, therefore we project the 3D feature points onto

the X-Y plane, then a part of footprint is obtained us-

ing RANSAC algorithm. We consider the footprint to

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

512

be regular. Finally, we can generate the whole foot-

print. In real environment, we have another informa-

tion acquisition, the existing GIS data, which can sup-

ply useful information for footprint generation.

3.3.2 Facades Modeling

Since we have discarded the noisy feature points, and

both the feature points and footprint have been di-

vided after the partition procedure in Section 3.2.2,

the last reconstruction step is to ﬁnish facades model-

ing. We use (Liu et al., 2011) method to achieve this

step.

We assume that the bottoms of all facades are

in the same plane and all facades are vertical to the

ground plane. With these hypotheses we need to re-

cover 3 parameters for each facade: left boundary and

right boundary location and height.

Based on section 3.2.2, the vector data of footprint

are divided into different facades. The facade enables

us to estimate the left boundary and right boundary

location, the facade height is the unique parameter

calculated. We consider there is a series of buffer re-

gions (BF) surrounding the footprint limits (Figure 2

(a)), therefore, we use the value of the highest point in

each buffer as the height of corresponding facade, as

shown in Figure 2 (b). After obtaining the max value,

several geometric planes can be modeled as facades.

Figure 2: Reconstruction process: (a) buffer regions sur-

rounding footprint (b) max height point extraction.

3.4 Updating Process

In this section, we describe the back-projection of the

footprint and 3D model onto video. User knowledge

enables us to achieve this stage easily and we update

the GIS data interactively.

3.4.1 Matching Footprint and 3D Model on

Video

For matching, we should calculate the extrinsic cam-

era parameters.

Then the correspondence between the camera co-

ordinate system and the world camera coordinate sys-

tem is established. We have calibrated the camera

before the extraction process, and have obtained the

intrinsic camera parameters.

3.4.2 Footprint Updating

We can know if the footprint exists in GIS from user

knowledge. If it does not exist, it is added into the

GIS. If it exists an old one, we need to compare both

of them. After 2D footprint and 3D model back-

projected on the video, we can conﬁrm which foot-

print is correct using the user knowledge and update it

to replace the old one. This interactive process makes

the footprint more accurate.

4 RESULTS

We have implemented our system and tested it on

street-view videos captured in town.The following re-

sults have been implemented in C++ code and on a

PC with Intel Core i7-870 CPU processor, NVIDIA

Quadro FX 580 graphics card. We tested our algo-

rithm on a video of ”building” model, and it ran well

from the extraction to the reconstruction (Figure 3).

Figure 4 shows another experiment. The video is

acquired by webcam Logitech QuickCam Pro 9000,

ﬁxed on a vehicle. The three bottom ﬁgures and the

top-right ﬁgure show us the whole pipeline of our al-

gorithm. The video consists of 212 frames lasting

about 9 seconds. At the beginning of the extraction

process, we pause the video every 0.5 second. Then

we use the RANSAC algorithm. Each iteration selects

randomly 3 feature points, and distance threshold is

0.05 measured in world coordinate system. Usually,

we have 3000 to 4000 feature points, and choose the

probability, that at least one of the random samples of

points doesn’t include any outlier, as 0.99.

In this experiment, due to the occlusions in out-

door video, footprint generation is in progress. As

for GIS data, we use ESRI Shapeﬁle and the building

layer of Nantes. Then we load and update it through

the OGR library, which is an open source code to deal

with GIS vector data, as shown in three top ﬁgures in

Figure 4.

5 CONCLUSIONS

We have proposed a GIS updating approach using

a video captured along the street, producing the ge-

ometric facades model and updating GIS data ﬁ-

nally. This approach enables us to update footprint

and add information to GIS. This method depends

on some user knowledge to make some interactive

operations. But the whole process reconstruct a 3D

model perfectly only with our box building, however

with outdoor real building, the facades reconstruction

A GIS DATA UPDATING METHOD WITH VIDEO

513

Figure 3: One updating GIS data process with a ”building” model.

Figure 4: Another updating GIS data process: the footprint of new building doesn’t exist in old GIS data.

and footprint generation are not yet available. These

works are in progress. There are several limitations,

such as it is not yet a real-time process and the updat-

ing process in outdoor environment faces the prob-

lem of occlusions. For our future work, we will fol-

low three directions. Firstly, we will make it work in

outdoor environment. Secondly, we will improve the

updating process, for example, the correction part. Fi-

nally, we will update 3D GIS by computing the height

of building and insert this information into GIS data.

REFERENCES

Den

egre, J. and Salg

e, F. (2004). Les syst

emes

d’information g

eographiques. Presses Universitaires

de France, 2nd edition.

Fischler, M. A. and Bolles, R. C. (1981). Random sample

consensus: a paradigm for model ﬁtting with appli-

cations to image analysis and automated cartography.

Commun. ACM, 24:381–395.

Goesele, M., Snavely, N., Curless, B., Hoppe, H., and Seitz,

S. (2007). Multi-View Stereo for Community Photo

Collections. In IEEE 11th International Conference

on Computer Vision(ICCV) 2007, pages 1–8.

Heipke, C., Woodsford, P. A., and Gerke, M. (2008). Updat-

ing geospatial databases from images. In Advances in

photogrammetry remote sensing and spatial informa-

tion sciences 2008 ISPRS congress book, pages 355 –

362. CRC.

Kelly, T. and Wonka, P. (2011). Interactive architectural

modeling with procedural extrusions. ACM Trans.

Graph., 30:14:1–14:15.

Klein, G. and Murray, D. (2007). Parallel tracking and

mapping for small AR workspaces. In Proc. Sixth

IEEE and ACM International Symposium on Mixed

and Augmented Reality (ISMAR’07), Nara, Japan.

Li, Y., Zheng, Q., Sharf, A., Cohen-Or, D., Chen, B., and

Mitra, N. J. (2011). 2d-3d fusion for layer decomposi-

tion of urban facades. In IEEE International Confer-

ence on Computer Vision (ICCV), Barcelona, Spain.

Liu, R., Servi

eres, M., and Moreau, G. (2011). Facade mod-

eling from a ground view video with map constraints.

In IAPR Machine Vision Applications, Japan, Nara.

uller, P., Wonka, P., Haegler, S., Ulmer, A., and Gool,

L. (2006). Procedural modeling of buildings. ACM

Transactions on Graphics, 25(3):614–623.

Pollefeys, M., Nist

er, D., Frahm, J. M., Akbarzadeh, A.,

Mordohai, P., Clipp, B., Engels, C., Gallup, D., Kim,

S. J., Merrell, P., Salmi, C., Sinha, S., Sinha, S., Tal-

ton, B., Wang, L., Yang, Q., Stew

enius, H., Yang, R.,

Welch, G., and Towles, H. (2008). Detailed real-time

urban 3D reconstruction from video. International

Journal of Computer Vision (IJCV), 78(2-3):143–167.

Steinocher, K. and Kressler, F. (2006). Eurosdr change de-

tection report. EuroSDR Ofﬁcial Publication Series,

50:111–182.

Zebedin, L., Bauer, J., Karner, K., and Bischof, H. (2008).

Fusion of feature-and area-based information for ur-

ban buildings modeling from aerial imagery. Euro-

pean Conference on Computer Vision(ECCV), pages

873–886.

GRAPP 2012 - International Conference on Computer Graphics Theory and Applications

514