Information Technology Reference
In-Depth Information
Fig. 11.8 The left and right image of the stereo camera and 3D range image a Left image; b
Right image; c 3D range image
In the proposed method, the initial size of the window is 16 9 12, the shift size
is 4 pixels, and the window size is increased by 10 pixels. The sampling times are
1,637,305 and the TH1 value is 100. The parameter is voted on only if the angle of
two normal vectors is less than 10. Also, the weight of the candidate normal
vector is set to a higher value when it has better planarity. The following rule
applies: voting count = current voting count + (angle threshold - angle of two
normal vectors). For example, if the angle is 0, which is the best planarity, it is
voted as having the highest value (10) and if the angle is 9, which is the worst
planarity, it is voted as having the lowest value (1). The parameter value is voted
as the mean of the two normal vectors. The TH2 value for removing the plane data
is 0.4 in both the conventional RHT method and the proposed method. We can get
these threshold values according to our experiments.
Figures
11.9
-
11.10
and Tables
11.2
-
11.3
show the comparison results of the
conventional RHT method and the proposed method. The top images are obtained
from the bumblebee stereo camera. These are the left, right, and 3D range images,
respectively. The middle ones are the results of plane detection, which are rep-
resented by the segmentation image and Hough parameter space. The colors of the
middle left images denote the result of the detected planes and the right graphs
show the result of the voting count of the Hough parameter space during each
iteration of IRHT. The graph of the Hough parameter space only represents the h
and u values and not the q value. The bottom table represents the q, h, u values for
plane detection.
Figure
11.9
and Table
11.2
show an experiment for one plane detection on an
uneven flat surface. The proposed method detected the two planes instead of one
Search WWH ::
Custom Search