Stereoscopic determination of terrain elevation

• US 5,309,522 A
• Filed: 06/30/1992
• Issued: 05/03/1994
• Est. Priority Date: 06/30/1992
• Status: Expired due to Fees

• Current Assignee: Environmental Research Institute of Michigan
• Original Assignee: Environmental Research Institute of Michigan
• Inventors: Dye, Robert
• Primary Examiner(s): NOT, DEFINED
• Assistant Examiner(s): Vigushin, John B.
• Application Number:
  US07/906,554
• Time in Patent Office: 672 Days
• Field of Search: 382/39, 382/41, 382/49, 356/2, 356/12, 358/88, 358/133
• US Class Current: 382/154
• CPC Class Codes: G01C 11/06 by comparison of two or more pictures of the same area
• CPC Class Codes: G06T 17/05 Geographic models
• CPC Class Codes: G06T 2207/10012 Stereo images
• CPC Class Codes: G06T 7/593 from stereo images
• CPC Class Codes: G06V 20/13 Satellite images
• CPC Class Codes: G06V 2201/12 Acquisition of 3D measurements of objects
• CPC Class Codes: H04N 13/221 using the relative movement between cameras and objects
• CPC Class Codes: H04N 2013/0081 Depth or disparity estimation from stereoscopic image signals

Abstract

Digitized stereo image pairs are analyzed by first aggregating the images to produce several levels of lower resolution images, bandpass filtering each of the images by applying a Laplacian filter to edge enhance the features, matching the features of low resolution images while generating a disparity image which contains elevation information, repeating the feature matching at successively higher resolutions and creating progressively improving disparity images, and extracting the elevation data from the final disparity image. The feature matching procedure uses an estimate of feature disparity to help locate a feature in one image which corresponds to a feature in another image and is applied iteratively at each level of resolution. Dilation of disparity data from matched points to neighboring unmatched areas produces a comprehensive disparity image which aids in more efficient feature location on the next pass.

Claims

1. The method of extracting elevation information form a pair of stereo images of terrain comprising:

reducing the resolution of each image of a pair of stereo images;

repeating the resolution reduction step to obtain several successively lower levels of image resolution;

enhancing edges in each level of image resolution by band-pass filtering the images;

matching features of the filtered image pair at the lowest level of image resolution by

estimating an average elevation as a disparity image,

steering a search for a feature in one image of a pair to match a feature found in the other image by using the disparity image to define the expected relative feature positions, and

searching for a matching feature in the said one image over a limited area around an expected position,

wherein the features have expected relative positions in the two images and elevation of a feature causes a disparity in expected positions;

determining the disparity for each matched feature and generating a disparity image from the disparity of matched features;

then matching features at successively higher levels of resolution using the disparity image generated at the next lower level of resolution to steer the search for a matched feature, and generating a new disparity image for each resolution level; and

extracting elevation information from the disparity image for the highest level of resolution.

2. The invention as defined in claim 1 wherein each step of reducing the resolution comprises aggregating the image by a factor of two.

3. The invention as defined in claim 1 wherein the band-pass filtering comprises applying a Laplacian filter to the images.

4. The invention as defined in claim 1 wherein the step of matching features comprises the step of:

iterating the search for matching features at the same resolution level using the most recently generated disparity image to steer the search, whereby the incidence of matched features increases with each iteration.

7. The invention as defined in claim 1 wherein the stereo images are formed by cameras at known heights and nadir points, and wherein the step of extracting elevation information from the disparity image at the highest level of resolution comprises the steps of:

establishing a reference datum;

determining the projected locations of a feature point on the reference datum as it is imaged at each camera location;

determining the projection coordinates of a feature point on the reference datum; and

calculating the elevation for the feature point on the basis of known camera heights and nadir points, the projection coordinates and the projected locations.

5. The method of extracting elevation information from a pair of stereo images of terrain comprising:

reducing the resolution of each image of a pair of stereo images;

repeating the resolution reduction step to obtain several successively lower levels of image resolution;

enhancing edges in each level of image resolution by band-pass filtering the images;

matching features of the filtered image pair at the lowest level of image resolution wherein the features have expected relative positions in the two images and elevation of a feature causes a disparity in expected positions;

determining the disparity for each matched feature and generating a disparity image from the disparity of matched features;

then matching features at successively higher levels of resolution using the disparity image generated at the next lower level of resolution to steer the search for a matched feature, and generating a new disparity image for each resolution level by filling image areas lacking a matched feature with the measured disparities of adjacent matched features to thereby approximate the disparity over the whole image; and

extracting elevation information from the disparity image for the highest level of resolution.

6. The invention as defined in claim 5 wherein the step of matching features at successively higher levels of resolution comprises the step of:

iterating the search for matching features at the same resolution level using the most recently generated disparity image to steer the search, whereby the incidence of matched features increases with each iteration.