Simplified image correlation method using off-the-shelf signal processors to extract edge information using only spatial data

• US 6,259,803 B1
• Filed: 06/07/1999
• Issued: 07/10/2001
• Est. Priority Date: 06/07/1999
• Status: Expired due to Term

• Current Assignee: The United States of America As Represented By The Secretary of Agriculture
• Original Assignee: the united states of america as represented by the secretary of the navy
• Inventors: Wirtz, Michael M., Ditzler, William R.
• Primary Examiner(s): Buczinski, Stephen C.
• Application Number:
  US09/337,219
• Time in Patent Office: 764 Days
• Field of Search: 382/103, 382/199, 701/223, 356/4.01
• US Class Current: 382/103
• CPC Class Codes: G06V 10/255 Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
• CPC Class Codes: G06V 10/7515 Shifting the patterns to accommodate for positional errors

Abstract

Provided is an approach to efficiently correlate a previously captured digitally created image to one provided in real-time. The real-time digitally created image is represented by the digitally processed image of just the edges of objects within a scene. This is accomplished via digital edge extraction and subsequent digital data compression, based on comparing only the spatial differences (e.g., range values) among pixels. That is, digital data representative of signal intensity are not used. An application is the efficient correlation of real-time digitally processed 3-D images generated from laser scans, in particular, scans of laser “radars” or LADARS. The process simplifies and improves on conventional techniques by iterating three sequential steps. A “hard” edge or a corner of an object is detected via a “corner-detector” algorithm that assigns a raw edge-strength value to each pixel in the image digitally created from the LADAR return. This is accomplished via a spatial numerical second-derivative operation performed on only the range value (distance from the LADAR to a specific location on an object scanned by the LADAR) assigned to each pixel'"'"'s neighboring pixels. Next, all edge-strength values greater than a pre-specified value representative of that obtained by a laser (light) reflection from a right angle (90°), i.e., a hard edge or a corner, are reset to the pre-specified edge-strength value. If this second step were omitted, the image correlation process would be dominated by large edge-strength values representative of actual discontinuities in an adjacent pixel'"'"'s range value, such as those produced when a tall object (building, vehicle, or promontory) shadows the area behind it from a laser scanning it obliquely. Finally, when a discontinuity in range values between two adjacent pixels is detected, the edge-strength value of the pixel representing the larger range value is set to zero. This last step validates the perspective transformation from an oblique view to an overhead view by avoiding the designation of a strong “phantom edge” on the ground. This phantom edge results from the laser beam forming a “shadow” when obliquely radiating a tall vertical object. The data needed to accomplish this correlation are minimized for subsequent storage and manipulation by storing only the two end points of the straight lines digitally generated to depict the edges of objects within a scene. This simple method, using inexpensive commercial off-the-shelf signal processors, enables reliable real-time identification of objects by comparison to a previously obtained reconnaissance library of digital images that have been stored for the purpose of future targeting. To enable each object in this library of digital images to be efficiently stored in memory, each digital image is defined by a greatly reduced and then compressed data set. Using this method, object identification may be accomplished quickly and reliably during such time-critical events as the terminal portion of a guided weapon'"'"'s trajectory.

Claims

1. A method for designating a target comprising:

establishing a reference image template comprised of a first set of pixels that may be represented by a first data set of points in at least one plane;

capturing said first data set in a manner that defines contrast boundaries of said reference image template and orientation thereof;

identifying data obtained from only those said pixels determined to be maxima edge pixels as a limited data set of said first data set of said pixels;

tracing edges along said limited data set to establish a data set of edges;

compressing said second data set by fitting straight line segments, each having first and second end points, to said data set of edges and storing in memory only said end points;

converting said data set of edges to at least one second data set to be represented in three-dimensional space;

providing a real-time image comprising a second set of pixels;

establishing a range value for each said pixel in said real time image;

based only on said range values, computing a gradient based edge-strength value for each said pixel in said real time image;

transforming said real time image perspective and range to that of said reference image template;

scaling said real time image to that of said reference image template;

correlating said scaled real time image to said reference image template;

designating a target within said reference image template; and

overlaying said designated target within said reference image template upon said real time image,

wherein an enhanced image of said designated target is provided for use in real time applications.

2. The method in accordance with claim 1 wherein said data in said data sets are digital.

3. The method in accordance with claim 1 wherein said reference image template is produced in advance of a requirement for use in real time.

4. The method in accordance with claim 1 wherein said reference image template is produced from reconnaissance images.

5. The method in accordance with claim 1 wherein three-dimensional data representative of said reference image template is used to connect said end points in three-dimensional space.

6. The method in accordance with claim 1 wherein said end points of at least one said line segment may be assigned a third dimension of zero.

7. The method in accordance with claim 1 wherein said real time image is provided by a LADAR system.

8. The method in accordance with claim 1 wherein, a raw edge-strength value may be assigned to each said pixel to facilitate identification of a structural corner in said real time image, wherein all said edge-strength values greater than a value representative of a reflection from a corner of an object are reset to said value representative of a reflection from a corner of an object, and wherein upon detecting a range discontinuity between two adjacent said pixels, said edge-strength value of said pixel with the highest range value is set to zero.

9. A image correlation system comprising:

a memory device for storing at least one template of at least one image of at least one pre-specified target, having edges at least one of which said edges is pronounced,

wherein said at least one image of said at least one pre-specified target is captured in said at least one template by at least one tracing along said pronounced at least one edges;

an imaging device for capturing in real time video images, comprising pixels, of at least one remote object;

a computer for processing data on spatial relationships,

wherein said processing identifies said remote objects by determining a range value for each of said pixels associated with said at least on remote object and computes at least one gradient based edge-strength value for each of said pixels associated with said at least one remote object;

at least one algorithm that;

resets at least one edge-strength value that is greater than a first value, said first value representative of a reflection from a corner, to said first value,

resets to zero said range value of a first at least one said pixel that is adjacent at least one second said pixel, said second at least one pixel having a shorter range value than said first at least one pixel,

wherein at least one structural edge of said at least one remote object is identified by tracing said pixels associated with said first value to establish an outline of said at least one object described by said at least one structural edge;

correlates said outline of said at least one object with said at least one image of said at least one pre-specified target in said stored template; and

overlays said at least one pre-specified target in said stored template on said real time image of said at least one object,

wherein said overlaying process enhances recognition of said pre-specified target in said real time image.