TS
4
Ad
MI
LC-
IN-
lO-
ive
VÍs
)7.
LO-
10-
Sergey Zheltov
METHOD OF 3D-OBJECT DETECTION BASED ON ORTHOPHOTO DIFFERENCE ANALYSIS
Sergey ZHELTOV, Alexander SIBIRYAKOV
State Research Institute of Aviation Systems (GosNIIAS), Moscow, Russia
zhl@ gosniias.msk.ru, sibiryakov ? mtu-net.ru
Working Group IC-13
KEYWORDS: Object detection, orthophoto difference, surface model, matched filter
ABSTRACT
In the paper we propose the method of object detection based on analysis of orthophoto images difference. Object is
registered stereoscopically on smooth surface of arbitrary shape. Orthophoto images for some basic plane are generated
from left and right images of the stereopair. Orthophoto generation is based on a priori surface model knowledge. In
case of object presence in the scene the orthophoto difference image contains a characteristic structure, which is
detected by the matched filtration.
1. INTRODUCTION
Problem of detection of small 3D-objects included in scene of observation often occurs in various machine vision
applications. Typical examples of this problem are obstacle detection for mobile robots or vision-guided navigation
[M.Bertozzi and A.Broggi, 1997].
In the presented work the problem is considered when probable detecting object is located on relatively smooth curved
surface like car road or way of mobile robot tracking. Analytical surface model might be known a priori with some
accuracy. Another way of background surface model acquisition is least square method of fitting to the data set of
points, which are exactly known to be not belonging to the object under detection.
In the case under considering object detection should be done stereoscopically, since in the opposite case it is
impossible to recognize situations when the object stands on the surface or is drawn on the surface. For stereoscopic
system its relative and interior orientation should be known.
The basic idea of the detection method consists of the following. If an analytical model of a surface is known,
orthogonal projections of the given surface to some convenient plane with use of left and right images of the stereopair
are created. The calculation of a difference of such projections results in appearance of characteristic geometric
structures in neighborhoods of 3D-objects not belonging to the given surface.
As orthogonal projections eliminate the distortions caused by surface irregularity the significant brightness variance is
appeared in difference image only in 3D-object neighborhood. In case of simple objects with straight-line edges the
precise 2D-corner shaped structures are appeared. Angle value and shape of corner are known functions of object
position do not depending on the object form. In case of objects with the more complex form (e.g. with plenty of edges)
a combination of angular structures dependent only on the object position is appeared on the difference image.
Therefore detection algorithm based on angular structures optimal matched filtration is the best in the given problem.
Thus, the work shows that the problem of 3D-object detection can be reduced to a problem of detection of 2D-
structures with the predicted properties on a synthetic image.
In the work presented the utmost accuracy of the suggested detection algorithm is analyzed depending on both object
elevation under surface and accuracy of relative and interior orientation. It is shown that using of standard video
cameras with the basis of the stereo system about Im provides the possibility of small object detection down to 5cm in
height on distance up to 80m. The results of using the method in the problem of road obstacle detection are shown.
International Archives of Photogrammetry and Remote Sensing. Vol. XXXIII, Part B3. Amsterdam 2000. 1041