A chip visual localization method based on vision for IC packaging equipment was developed to satisfy the high
precision requirement in the chip inspection in this paper. The method is based on the combination of Hough transform
and prior geometric knowledge of chip. First, a straight line was extracted to match chip image.Then the conversion
function of the chip image was deduced based on the straight line endpoint coordinates. Because the relative position
coordinates of chip corner are known, the other two endpoints of chip in image can be deduced with substituting the
known model coners coordinates into the matching conversion function. Finally, the chip orientation is achieved by
linking the corners coordinates. Experimental results have shown that the the chip can be orientated accurately using this
Current research on computer vision often needs specific techniques for particular problems. Little use has been made of high-level
aspects of computer vision, such as three-dimensional (3D) object recognition, that are appropriate for large classes of problems and
situations. In particular, high-level vision often focuses mainly on the extraction of symbolic descriptions, and pays little attention to
the speed of processing. In order to extract and recognize target intelligently and rapidly, in this paper we developed a new 3D target
recognition method based on inherent feature of objects in which cuboid was taken as model. On the basis of analysis cuboid nature
contour and greyhound distributing characteristics, overall fuzzy evaluating technique was utilized to recognize and segment the target.
Then Hough transform was used to extract and match model's main edges, we reconstruct aim edges by stereo technology in the end.
There are three major contributions in this paper. Firstly, the corresponding relations between the parameters of cuboid model's
straight edges lines in an image field and in the transform field were summed up. By those, the aimless computations and searches in
Hough transform processing can be reduced greatly and the efficiency is improved. Secondly, as the priori knowledge about cuboids
contour's geometry character known already, the intersections of the component extracted edges are taken, and assess the geometry of
candidate edges matches based on the intersections, rather than the extracted edges. Therefore the outlines are enhanced and the noise
is depressed. Finally, a 3-D target recognition method is proposed. Compared with other recognition methods, this new method has a
quick response time and can be achieved with high-level computer vision. The method present here can be used widely in vision-guide
techniques to strengthen its intelligence and generalization, which can also play an important role in object tracking, port AGV, robots
fields. The results of simulation experiments and theory analyzing demonstrate that the proposed method could suppress noise effectively, extracted target edges robustly, and achieve the real time need. Theory analysis and experiment shows the method is
reasonable and efficient.
Object-identification using edge extraction techniques from background function in uncontrolled lighting environments
containing more object pose information and have many applications. In order to depressing noise, identify aim body
robustly and rapidly, in this paper, We take cuboid as model and present a new strategy for edge extraction and object-identification
based on object inherent features. This strategy includes the following steps. Firstly, pre-processing is
applied to the raw image, in which Canny operator was used to extract edges pixels, then, image was divided into a grid
of overlapping windows and noise was suppressed by regression grid windows in which the number of pixels is less than
a threshold. Secondly, as model contour's geometry characters known already, the cuboids upright edges was used as
their existence evidence to estimate model's existence area and so the lines failed spatial constraints are eliminated, then,
object edges was extracted within the finite ranges of orientation in Hough transform space. Thirdly, the intersections of
the component extracted edges are taken, the candidate edges extraction and matches was assessed based on the
intersections, rather than the component extracted edges. After a series of matching tests the aim body is extracted.
The proposed method makes three major contributions. Firstly, on the base of study the correspondence between model's
boundary edges parameters in image space and Hough space we extract edges in finite area in Hough transform space,
the aimless computations and searching is reduced greatly, its efficiency improved. Secondly, as Canny operator can
extract aim lines with single pixel width, the edges extraction strategy of combining Canny operator with Hough
transform extractor could avoid error impact of edges pixels numbers to Hough extractor. Thirdly, after fusion model's
knowledge in image space, Hough space, global space, learning from others strong points to offset one's weakness, we
extract model's edges from complex noise background without regarding to regression caused by the errors due to
spurious or missing pixels because edge extraction is imperfect for real images.
The results of experiments demonstrated that the proposed method could suppress noise effectively, identified and
extracted target from complex backgrounds robustly. This new strategy may have potential application in visual servo,
object tracking, port AGV and robots fields etc.