Information Technology Reference
In-Depth Information
is then computed by determining the maximum distance between the detected
lane marker locations and that of the ground truth. The ground truth data is
also transformed to the IPM domain using Eq. (1) and (2) to allow the accurate
computation of these distances. In the bird's-eye view, the inter-pixel distances
have linear correspondences in the world; as a result, the distances computed in
any portion of the image can be easily mapped to a physical distance.
The distances are computed at one foot intervals up to 30ft ahead of the
vehicle. The error is determined by calculating
W
2
λ ( i,f ) = max (
|
Gt ( i,f )
X ( i,f ) |−
, 0) s.t. i
[0 , 30]
(7)
E(f) =
λ =max
i
λ ( i,f )
(8)
where Gt ( i,f ) is the ground truth location of the lane marker and X ( i,f ) is the
detected lane location in frame f at a distance of i feet ahead of the car. W
is an interval around the ground truth locations and is set to the equivalent of
8 inches in the IPM image. This value is chosen as the mean of the widths of
normal and wide lane markers based on the specifications of the Federal Highway
Administration (FHA) [13]. Consequently, lane marker estimates that fall within
the interval specified by W are categorized as having no error. As a result, the
error in each frame, E(f) is computed as the L-Infinity Norm of the λ values.
This idea is illustrated in Fig. 8 where the green line marks the ground truth,
the blue line is the lane marker estimation using the proposed lane detector, and
λ ( i,f ) is the offset measured at specific distances ahead of the vehicle.
0ft
30ft
(0,f)
(4,f)
(8,f)
(30,f)
}W
Fig. 8. Calculating errors using λ distances
8 Results and Analysis
The following rules were used to quantify the results into the different categories:
1. A correct detection occurs when less than 2
λ distances are greater than 0.
2. A missed detection occurs when more than 2
λ distances are greater than 0.
Search WWH ::




Custom Search