The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. XXXVII. Part Bib. Beijing 2008 
Hypothesis Analysis (MHA) method is used to link the LLSs 
into long lane lines 3D model (see section 6). Each lane line is 
then classified into a lane line type, for instance “white dash 
line”. The classification is based on the analysis of the LLSs’ 
features within the lane line. A continued lane line may be of 
different types in difference sections, for example “white dash 
line” may change to “white solid line” near traffic light 
intersections. A decision filtering method is used to find the 
type-changed points (see section 7). Figure 4 shows the 
flowchart of ARVEE framework. 
Figure 4: ARVEE Workflow 
4. PITCH CORRECTED INVERSE PERSPECTIVE 
MAPPING (PCIPM) 
The inverse perspective mapping (IPM) can be used to simplify 
the process of lane detection. IPM essentially re-projects the 
images onto a common plane (road surface plane) and provides 
a single image with common lane line structure. As shown in 
Figure 5, direction OA is the optical axis of a given camera. IG 
is the ideal road surface plane. Assuming the vehicle is a rigid 
body, the angle between OA and IG can be estimated at system 
calibration stage. 
We denote the angle between the two vectors as Angle(-,-). The 
classical inverse perspective mapping assumes that the vehicle 
drives on a perfect flat plane IG, and Angle(OA, IG) is fixed, 
and the distance from camera to the road surface (denoted as H) 
is fixed; both Angle(OA, IG) and H are known. The IPM 
projects all the original images from different cameras onto the 
IG plane, and generates the 2D ortho-image on the road surface 
plane, as shown in Figure 5. The generated image is no longer a 
perspective image but a map. In Figure 5, M is a plane parallel 
to IG. The ideal IPM result can be a mapping from IG to any M, 
through the direction that perpendicular to IG. 
M 
Figure 5: Classical IPM 
The classical IPM is based on the flat road surface assumption. 
However, this assumption is not always valid in real world. 
There are several facts that invalidate the assumption. First, 
road surface is not always an ideal plane, in stead; a curved 
road surface is common in practice. As shown in Figure 6, 
given G is the true road surface, the classical IPM will project 
the on-road-surface point (a) to position (a); and this will 
cause distortion in the resulted IPM map. Second, due to the 
flexibility of tires and shock absorber, vehicle is not a rigid 
body, the ideal IPM is violated. Therefore, given the same road 
surface and the same vehicle position, the angle between OA 
and the road surface may still be different. Third, small bumps 
on the road surface may cause fluctuation of vehicle, and again, 
this will cause OA to change. 
In ARVEE, we introduce georeferencing information to 
overcome this problem. VISAT™ provides quite accurate 
measurement of the position of the body frame centre. All the 
cameras are fixed to the navigation body frame; and the 
relationship of all the sensors are accurately estimated during 
the system calibration. 
Figure 6: Pitch corrected IPM 
The trajectory of the survey provides a good estimate of road 
surface profiles, and this model is in earth mapping frame. 
Given the positions of the survey points as {p;}, the estimation 
of the trajectory at position i can be expressed as 
Ti = F{P i _ m ,...,P i _ v P i ,P M ,...,P i+n ) (l) 
where F is a trajectory interpolation function, which takes the 
ordered position sequence and models the trajectory. Since the 
road surface behind the vehicle has no influence to the IPM, so 
m = 0. The road surface are suppose to be smooth, so n can be 
2, and F can, therefore, be defined as 
(2) 
n ~ 
7=1 
The road surface can be described as a general cylindrical 
surface along the trajectory. At each point on the trajectory, the 
roll angle between the road surface to the body frame 
coordinate is 0.0. With this assumption, we correct the classical 
IPM according to the local trajectory at each survey point. The 
proceeding processing stages use this pitch-corrected road 
517