Method of detecting moving objects in chronologically successive images

• US 6,069,918 A
• Filed: 09/11/1997
• Issued: 05/30/2000
• Est. Priority Date: 01/15/1996
• Status: Expired due to Term

• Current Assignee: Robert Bosch GmbH
• Original Assignee: Robert Bosch GmbH
• Inventors: Meyer, Michael, Hoetter, Michael, Rottmann, Frank
• Primary Examiner(s): RAO, ANAND SHASHIKANT
• Application Number:
  US08/913,892
• Time in Patent Office: 992 Days
• Field of Search: 348/169, 348/172, 348/700, 348/699, 348/170, 348/416, 348/415, 348/412, 348/413, 375/240.17, 375/240.18, 375/240.9
• US Class Current: 375/240.17
• CPC Class Codes: G06T 2207/10016 Video; Image sequence
• CPC Class Codes: G06T 7/231 using full search
• CPC Class Codes: G06V 10/42 Global feature extraction by analysis of the whole pattern, e.g. using frequency domain transformations or autocorrelation

Abstract

A method for detecting moved objects in chronologically successive pictures is disclosed in which a current picture is divided into picture blocks and in each picture block a change of a picture signal of the current picture from that of a corresponding picture block of a first reference picture preceding the current picture is as certained , To detect the change the difference between the picture signals of the current picture and tho se of a second reference picture without moved objects is first compared pixel-precisely with a first threshold. If the first threshold is exceed, an object-oriented analysis of object size and/or shape takes place in which the picture signals corresponding to the moved object are compared with picture signals of a corresponding region of the first reference picture. For at least one moved object of the current picture that undershoots a size threshold, a motion vector is estimated. If the magnitude of the estimated motion vector exceeds a second threshold, the associated object is monitored chronologically and if a third threshold is exceed, then the object is immediately detected as a moved object. For the chronologically monitored objects, based on a third reference picture corresponding to a previous current picture object motion is estimated to predict a new object position in the current picture.

Claims

1. A method for detecting moved objects in chronologically successive pictures, wherein for block-oriented analysis a current picture is divided into picture blocks, for each picture block one picture signal of the current picture is compared with a picture signal of a corresponding picture block of a first reference picture that precedes the current picture in the chronological succession of pictures,

in that in particular a picture signal change and/or a textural change is ascertained, characterized in that for detecting a change, the difference between the picture signals of the current picture and the picture signals of a second reference picture without moved objects is compared pixel-precisely with a predetermined first threshold;

that if the predetermined first threshold is exceeded, then in an object-oriented analysis the size and/or shape of an object is ascertained and the picture signals that represent the moved object are compared with the picture signals of the corresponding picture region of the first reference picture, in particular by ascertaining a picture signal change and/or a textural change;

that for at least one moved object of the current picture that undershoots a predetermined first size threshold, a motion estimate is made by association with a corresponding picture region of the first reference picture and formation of an associated motion vector;

that if a predetermined second threshold is exceeded by the amount of the estimated motion vector, the object is delivered for chronological monitoring; and

that if a predetermined third threshold is exceeded by the amount of the estimated motion vector, the object is detected as a moved object.

2. The method as defined in claim 1, wherein for at least one chronologically monitored object, a motion estimate is performed, on the basis of a third reference picture that corresponds to a previously used current picture, forming an associated motion vector and with corresponding prediction of a new object position for the current picture;

that in the pixel-precise detection of change, the difference between the picture signals at the predicted object position of the current picture and the corresponding picture signals of the second reference picture are compared pixel-precise with the predetermined first threshold;

that if the predetermined first threshold is exceeded, then with an object-oriented analysis the size and/or shape of an object is ascertained and the picture signals that represent the monitored object are compared, in an object-oriented way, with the picture signals of the corresponding picture region of the first reference picture by ascertaining the textural change;

that an association of the monitored object with the object of the current picture is carried out in the current picture by superimposing on the object position predicted by the motion estimate a corresponding object position ascertained in the object-oriented analysis;

that the motion of the monitored object is ascertained as a function of time by adding together motion vectors thus far; and

that if the predetermined third threshold is exceeded by the amount of the added-together motion vector, the monitored object is detected as a moved object.

3. The method as defined in claim 1 or 2, wherein the object-oriented analysis is performed only in those picture regions in which moved objects have been found by the block-oriented analysis and/or in which the motion estimate is performed for at least one chronologically monitored object.

4. The method as defined in claim 1, wherein for the monitored object, the mean object-oriented textural change over time is ascertained, and, if a predetermined fourth threshold for the mean object-oriented textural change over time is undershot, the monitoring of the object is discontinued.

5. The method as defined in claim 4, wherein to ascertain the mean object-oriented textural change over time, quadratic deviations of textures of the picture blocks that include at least part of the object from a chronological mean value of the textures are ascertained in the block-oriented analysis and are added together, weighted to form weighted sums, the weight of the summand being selected to be proportional to the surface area of the picture block by which the picture block covers the object;

and the weighted sums are added together in object-oriented fashion over time and are divided by the chronological sum of weighted sums of the chronological variance of the textures also ascertained in the block-oriented analysis.

6. The method as defined in claim 2, wherein, if the predetermined second threshold is undershot by the amount of the added-together motion vector of the monitored object, the monitoring of the monitored object is discontinued.

7. The method as defined in claim 2, wherein, if the predetermined second threshold is exceeded by the amount of the added-together motion vector of the monitored object, the monitored object is detected as a moved object, in the event that in the most recently performed motion estimation it was not possible to ascertain a reliable motion vector.

8. The method as defined in claim 2, wherein, if the predetermined second threshold is exceeded by the amount of the added-together motion vector of the monitored object, the monitored object is detected as a moved object, in the event that no association of the monitored object with an object in the current picture is possible.

9. The method as defined in claims 1 or 2, wherein objects whose size is below a predetermined second size threshold and objects whose shapes do not meet predetermined requirements are not detected as moved objects, and an object-oriented comparison of their picture signals is not performed.

10. The method as defined in claims 1 or 2, wherein further comprising tripping an object alarm for moved objects that are detected.

11. The method as defined in claims 1 or 2, further comprising discontinuing existing monitoring of moved objects that are detected.

12. The method as defined in claims 1 or 2, wherein, for the motion estimate within a predetermined search range for possible estimated motion vectors, a quality criterion for each of the possible estimated motion vectors is ascertained;

the motion estimate is evaluated as reliable if the estimated motion vector having the best quality criterion has a markedly better quality criterion than the other possible estimated motion vectors; and

the motion estimate is otherwise evaluated as unreliable.

13. The method as defined in claim 1, wherein object motion is ascertained by means of a displacement in the pixels contained in an object mask.

14. The method as defined in claim 13, wherein the object mask is obtained by means of the pixel-precise detection of change, and every coherent picture region detected as having changed is characterized with an identification code.

15. The method as defined in claim 2, wherein object motion is ascertained by means of a displacement in the pixels contained in an object mask;

for finding an association of the monitored object with an object of the current picture, the object mask of the previously used current picture is superimposed on the object mask of the current picture, taking into account the predicted object motion, and for an unambiguous association circumscribing rectangles of the corresponding objects must overlap.

16. The method as defined in claim 1, wherein the size of the moved object is ascertained in weighted fashion as a function of the picture position.

17. The method as defined in claim 1, wherein the first reference picture is copied into the second reference picture if no object is being monitored and all the objects in the current picture are detected as nonmoving objects.

18. The method as defined in claim 1, wherein the picture blocks having objects that are executing a slight motion are copied out of the first reference picture into a buffer memory, and after a change in the first reference picture the picture blocks having the objects with slight motion are copied back out of the buffer memory again into the first reference picture.

19. The method as defined in claim 1, wherein the first reference picture is overwritten with the current picture if no moved object or object characterized by an unreliable motion estimate is present;

and otherwise the first reference picture is overwritten with a subsequent current picture.