Image analyzer for processing multiple frames of image data

• US 4,464,789 A
• Filed: 05/14/1982
• Issued: 08/07/1984
• Est. Priority Date: 05/19/1980
• Status: Expired due to Term

• Current Assignee: Environmental Research Institute of Michigan
• Original Assignee: Environmental Research Institute of Michigan
• Inventors: Sternberg, Stanley R.
• Primary Examiner(s): Boudreau, Leo H.
• Application Number:
  US06/378,358
• Time in Patent Office: 816 Days
• Field of Search: 358/105, 358/107, 358/108, 358/109, 358/103, 343/5 MM, 343/7.7, 364/300, 356/388, 382/27, 382/49, 382/48
• US Class Current: 382/291
• CPC Class Codes: G06T 5/20 using local operators

Abstract

A method of analyzing multiple frames of image data wherein each frame is represented by a matrix of pixels fed serially from a source for processing. A first frame of pixels is transformed in at least one serial neighborhood transformation stage to detect spatial locations associated with given objects in the frame. A second frame of pixels is subsequently transformed to detect the spatial locations associated with the objects in that frame. Then, the differences between the transformations of the first and second frames are analyzed to detect changes in the number of and/or location of the objects between frames.

Claims

1. A method of analyzing multiple image frames, each frame being represented by a matrix of pixels fed serially from a source for processing, said method comprising:

transforming a first frame of pixels in at least one serial neighborhood transformation stage to detect spatial locations associated with a given object therein;

retaining indicia of the transformation of the first frame associated with the spatial locations of said object;

subsequently transforming a second frame of pixels in at least one serial neighborhood transformation stage to detect spatial locations associated with said object therein; and

analyzing the difference between the transformation of the first and second frames to thereby detect changes of said object between frames.

2. The method of claim 1 wherein each pixel is represented by a multibit binary data signal and each frame is transformed by a pipeline of individually programmable stages, said method further comprising the step of feeding back selected bits of each transformed pixel of the first frame associated with the spatial location of the detected object thereon from the output to the input of the pipeline.

3. The method of claim 2 which further comprises the steps of:

combining the selected feedback bits of the transformed first frame of pixel data with bits corresponding to the same pixels in the second frame; and

shifting said combined bits through the pipeline of stages, with said feedback bits being passed unmodified through selected stages of the pipeline.

4. A method of automatically detecting changes in particular objects between multiple frames of a scene, said method comprising:

scanning the scene to provide a first frame of serial pixel data;

subsequently scanning the scene to provide a second frame of serial pixel data;

using at least one serial neighborhood transformation stage to detect the spatial locations of a particular object in the scene by transforming the pixels in the first frame;

combining data associated with said transformed pixels with corresponding pixels in the second frame;

transforming the pixels in the second frame in at least one serial neighborhood transformation stage to detect the spatial locations of said object therein while passing the previously transformed data unmodified through the stage; and

comparing the differences between the transformations of the first and second frames to detect for changes of the object therebetween.

5. The method of claim 4 wherein said transformations are performed in a pipeline of substantially identical individually programmable stages.

6. A method of automatically detecting changes in particular objects between plural image frames taken of a generally common scene, each frame being represented by a matrix of pixels, said method comprising:

using neighborhood transformations to transform the matrix of a first frame to detect spatial locations within the matrix associated with a given object;

using neighborhood transformations to transform the matrix of a second frame to detect spatial locations within that matrix associated with said given object; and

analyzing the transformations of the first and second frames to thereby detect changes of said object between frames.

7. The method of claim 6 wherein said matrices are fed serially from a data source and are transformed in at least one serial neighborhood transformation stage.