Boots – shoes – and leggings
Patent
1993-01-29
1995-07-11
Ramirez, Ellis B.
Boots, shoes, and leggings
364516, 382191, 382199, G06K 920, G06K 1500
Patent
active
054327121
DESCRIPTION:
BRIEF SUMMARY
FIELD OF THE INVENTION
The invention relates to machine vision stereo matching, i.e., the matching of features between two different images of a scene.
BACKGROUND
The matching of features in images representing a scene viewed from at least two different viewing positions is a major problem in stereo vision in the art of machine vision systems. To date, this problem has been approached by applying a set of constraints so that areas or features or both in two or more images are matched. This approach has met with varying degrees of success because there is no general solution to the problem and a set of constraints applied to one scene may not be appropriate to other scenes. Improperly applied constraints will lead to an interpretation of the images which does not correspond fully with the scene.
There is a reduction of one dimension when a three-dimensional (3D) scene is captured in a two-dimensional (2D) image. This dimensional reduction results in a loss of information about the scene because a point in the 2D image will correspond to a line in the 3D scene. Thus, the process of 2D image formation can be regarded as a many-to-one mapping. I.e., many points in the 3D space of the scene will map onto a single point in the 2D space of the image. Thus, the inverse operation of mapping 2D image space points onto 3D scene space points is a one-to-many mapping.
This loss of information can be and is avoided by using at least one further image of the scene taken from a different viewing position and then identifying corresponding points in the plural images that relate to the same point in the 3D space of the scene. In this way, the process of mapping 2D image space points on to 3D scene space points is reduced to a one-to-one mapping. However, in adopting this approach a new problem is created, namely, the problem of identifying corresponding points in the plural images, i.e., the correspondence problem. For any one pixel in one image, there will be a large number of possibly corresponding pixels in the or each other image, the extreme being every pixel in the other image being a possible match with the pixel in the first pixel, and there is of course only one correct corresponding pixel.
SUMMARY OF THE INVENTION
The present invention provides a means by which the number of possible matches is reduced, thereby both reducing the processing time required to find a match and increasing the likelihood of making a good match.
To obtain accurately triangulated 3D measurements from stereo images, a wide baseline, i.e., the separation between cameras, or in the case of convergent stereo images, a wide viewing angle is required. The correspondence problem is more severe in wide baseline/angle stereo images because the changes between pairs of images can be very large and this in turn makes matching of images more difficult. These changes can be categorized into two types, geometric distortion and photometric distortion. Geometric distortion is the change in the geometrical description of a matching entity between images. Photometric distortion is caused by the change in orientation of a surface of an object in relation to the viewing point when a scene is viewed from a different viewing position. The degree of both types of distortion increases with increased baseline or viewing angle between the two viewing positions, although photometric distortion is less significant when a scene is evenly illuminated. In an embodiment of the invention to be described in greater detail hereinafter, a compromise is made so that the matching of images is possible and at the same time the recovered 3D measurements are accurate enough for subsequent use.
Most stereo matching algorithms recognize a common problem, namely, that matching is only possible for those image points with distinctive image characteristics. Image points within homogeneous regions cannot be matched. In general, pixel or area based matching is more prone to image noise, whereas feature based matching suffers from both photometric and geometric distortions. The
REFERENCES:
Medioni et al., Title "Segment-Based Stereo Matching", Computer Vision, Graphics, & Image Processing 31, 2-18 (1985).
Nasrabadi et al; Title "Stereo vision correspondence using a multichannel graph matching technique", 1988 IEEE pp. 1804-1809.
Pape, V.; Algovithm 562-Shortest Path Lengths [H]; ACM vol. 6, No. 3, Sep. 1980 pp. 450-455.
Axiom Innovation Limited
Ramirez Ellis B.
Shah Kamini S.
LandOfFree
Machine vision stereo matching does not yet have a rating. At this time, there are no reviews or comments for this patent.
If you have personal experience with Machine vision stereo matching, we encourage you to share that experience with our LandOfFree.com community. Your opinion is very important and Machine vision stereo matching will most certainly appreciate the feedback.
Profile ID: LFUS-PAI-O-508316