![](bg2.png)
a short time. The moving direction and the moving
distance are estimated from images of a backup cam-
era. The distance to the side wall is calculated only
while a vehicle goes straight. Since corresponding
between features is performed by a block matching
in the IPM image and the motion vector is estimated
from the peak of the histogram, our method is robust
to changes in the appearance of the texture on the side
wall that occurs when a vehicle moves along a road.
The remainder of this paper is organized as fol-
lows. In Sect.2, related works are reviewed briefly.
In Sect.3, the outline of the proposed method is de-
scribed. In Sect.4, the method to calculate the dis-
tance to the side wall is explained in detail. In Sect.5,
experimental results performed to simulation images
and several snow road scenes are discussed. Conclu-
sions are presented in Sect.6.
2 RELATED WORKS
Stereo vision is often used for measuring the dis-
tance to an object with cameras(Dhond and Aggar-
wal, 1989),(Hoff and Ahuja, 1989). Various meth-
ods which can acquire a dense depth map have been
proposed and used for automobile applications, such
as obstacle detection, road boundary detection and
so on(Einecke and Eggert, 2013),(Suhr and Jung,
2013),(M. Michael and Schlipsing, 2013),(C. Guo
and Naito, 2013),(M. Enzweiler and Franke, 2013).
A commercial car with the collision avoidance sys-
tem using a stereo camera has already been pro-
duced(Eyesight, 2013)(K. Saneyoshi and Sogawa,
1993)(Sogawa and Hanawa, 2002). Although stereo
vision is effective in distance measurement with a
camera, it requires higher cost than monocular vision
because it needs calibrated two cameras. Therefore,
there are few vehicles which have been equipped with
stereo cameras.
On the other hand, vehicles with a single camera
for rear view monitor or drive recorder are increas-
ing. Especially, vehicles with a backup camera for
parking support are rapidly increasing in Japan as the
car navigation system spreads widely. For this reason,
we proposes the method which measures the distance
to the snow side wall by a single camera, especially
a backup camera. Although a lot of methods have
been proposed for distance measurement with a sin-
gle camera, motion stereo is generally used in auto-
mobile applications(Huang, 1994)(A. Wedel and Cre-
mers, 2006)(A.J. Davison and Stasse, 2007). Motion
stereo needs to search corresponding points between
two frames taken at different points or it needs to track
feature points between frames. However, an image of
Figure 1: Snow wall of shoulder.
a backup camera has a severe distortion. In addition,
appearance of the texture on the side wall changes a
lot as the vehicle moves forward since the side wall
is parallel to the moving direction of the vehicle. In
addition, a lot of similar texture exist on the snow side
wall. Therefore, it’s difficult to estimate the distance
to the snow side wall stably by conventional methods.
3 OUTLINE OF THE PROPOSED
METHOD
Figure 2 shows the model of the driving environment
which the proposed method assumes. The distance
x between the on-vehicle camera and the side wall
along the road is calculated from the moving distance
of a vehicle and the movement of the side wall in the
IPM image when a vehicle moves forward. The pro-
posed method calculates the distance to the side wall
near a vehicle. Therefore, it is assumed that the side
wall consists of planes perpendicular to a road sur-
face.
Figure 3 shows the procedure of the proposed
method. At first, the motion of ego-vehicle is es-
timated from image sequences of a backup camera.
The motion vector is estimated from optical flows de-
tected on a road surface. Our method decides that
an ego-vehicle moves straight when the motion vec-
tor shows the upward direction in a certain period of
time. When a vehicle moves straight, the inverse per-
spective mapping (IPM) image is created and opti-
cal flows are detected in the IPM image by the block
matching. If the wall is close to a vehicle, the move-
ment of the wall in the IPM image is large as shown
in Fig. 4(a) and the magnitude of the optical flow is
also large. If the wall is far from a vehicle, the move-
ment of the wall in the IPM image is small as shown
in Fig. 4(b) and the magnitude of the optical flow is
also small. Then, a histogram whose bin is the mag-
nitude of the optical flow is created. The magnitude
SnowSideWallDetectionusingaSingleCamera
503