Eng, Swee Kheng
(2012)
Stereo Vision Using Compensation Technique For Improvement Of Distance Measurement Accuracy For Robot Arm Applications.
Masters thesis, Universiti Sains Malaysia.
Abstract
Two-Dimensional (2D) vision systems have been widely applied to various
automation fields in order to increase productivity. However, these 2D vision
systems are not able to retrieve the third dimension (depth) of the object. Therefore,
the most promising solution to this problem is to apply the Three-Dimensional (3D)
vision system to the automation fields. Among the 3D vision range finding method,
stereo vision is most effective way of realizing the 3D vision. This research uses a
pair of cameras as a stereo vision module to locate the object. The stereo vision
system, which consists of camera calibration, image rectification, image
segmentation, centroid computation and object localization is implemented. Image
segmentation is performed through a process of edge-based segmentation,
mathematical morphology and largest area selection. The x, y coordinates are given
by the centroid coordinates of the object in the left image and the z coordinates are
computed based on the obtained disparity value. The grayscale left and right images
are successfully aligned horizontally through the image rectification process. The
disparity error is minimized after the image segmentation process. The x coordinates
result for mean and standard deviation of error are -0.14 cm and 0.19cm, respectively.
The mean of error for y coordinates is -0.10 cm and the standard deviation of error
for y coordinates is 0.48 cm. The initial mean and standard deviation of error for z
coordinates are 0.37 cm and 0.42 cm, respectively. However, after compensation technique is performed, the mean and standard deviation of error for the z
coordinates are reduced to -0.12 cm and 0.16 cm, respectively. The 3-degree of
freedom robot arm can pick up a ball with the 3D location provided by the stereo
vision system.
Actions (login required)
 |
View Item |
