Detecting objects in images

Patent metadata

Field	Value
Publication number	US-2018267163-A1
Application number	US-201515761955-A
Country	US
Kind code	A1
Filing date	Sep 21, 2015
Priority date	Sep 21, 2015
Publication date	Sep 20, 2018
Grant date	—

Abstract

A method and a device for detecting at least one target in an image, wherein the image comprises a set of pixels with a magnitude assigned to each pixel is provided. The method comprises an iterative process until the K+1th target does not show a probability increase above a predetermined threshold value. The method is performed by creating a candidate free image, calculating, for the candidate free image, the probability of there being a target at each pixel, by using Bayes theorem, determining a location of the candidate target K+1 in the image, determining the probability that there is a target at the determined location, by determining the calculated probability of there being a target at the determined location. By performing the above, the most probable locations for targets in the image are located together with the probability that the location holds a true target.

First claim

1 . A method for detecting at least one target in an image (I), wherein the image (I) comprises a set of pixels with an magnitude assigned to each pixel, wherein the method comprises, for a number of candidate targets K≥0, and starting with K=0, performing the following until the K+1th target does not show a probability (P) increase above a predetermined threshold value:
creating (S1) a candidate free image by removing a footprint of the candidate target K from the candidate free image of the previous candidate target K−1; wherein if K=0, the candidate free image is equal to the image (I);
calculating (S2), for the candidate free image, the probability of there being a target at each pixel, by using Bayes theorem; wherein Bayes theorem is calculated using the number of targets K;
determining (S3) a location of the candidate target K+1 in the image (I) by:
identifying (S31) the location in the candidate free image having a maximum probability of there being a target at each pixel, the identified location being the location of the candidate target K+1;

determining (S4), for each candidate target from 1 to K, the probability (P) that there is a target at the determined location associated with that candidate target by determining (S41) the calculated probability of there being a target at the determined location;
repeating (S5) the above with K substituted with K+1.

2 . The method according to claim 1 , wherein Bayes theorem is calculated using a target probability distribution, a clutter probability for the magnitudes of the candidate free image, and a probability model for magnitude assuming that a target is present.

3 . The method according to claim 2 , wherein the target probability distribution is calculated using the number of pixels of the image (I), the number of targets and the number of pixels of a target footprint.

4 . The method according to claim 3 , wherein the target probability distribution is obtained by making an explicit assumption that the candidate targets K have a uniform distribution over the image (I).

5 . The method according to claim 3 , wherein target occurrence is weighted with the probability of a target to appear or not appear in different areas of the image (I).

6 . The method according to claim 2 , wherein probability model for magnitude assuming that a target is present is determined by assuming the appearance of clutter magnitude and target magnitude as independent occurrences and wherein the probability model for magnitude assuming that a target is present is the sum of all products of the probabilities of clutter magnitude occurrences and the probabilities of target magnitude occurrences in the image (I).

7 . The method according claim 6 , wherein it is assumed that the probability for target magnitude is uniform for all target magnitudes within a predetermined interval.

8 . The method according to claim 2 , wherein the image (I) comprises an image pair of one reference image (I r ) and one updated image (I u ), and each pixel of the image (I) comprises pairs of numbers with one element of the pair being the magnitude of the reference image (I r ) pixel and the other element the magnitude of the updated image (I u ) pixel, the reference image and the updated image being fully aligned over the same scene on a pixel to pixel basis.

9 . The method according to claim 8 , wherein clutter comprises the difference between the reference image (I r ) magnitudes and the updated image (I u ) magnitudes and clutter probabilities are due to random changes in magnitudes between the two images, and whereby the clutter probability for the candidate free image is determined by:
constructing a two dimensional histogram with respect to a reference image (I r ) magnitude and the magnitude of the magnitude difference between reference image (I r ) and updated image (I u );
wherein the clutter probability for the candidate free image is the conditional clutter probability determined by the two dimensional histogram.

10 . The method according to claim 9 , wherein the histogram bins are uniformly distributed on a logarithmic scale, chosen so that the population of the bins follows a monotonically decaying scale.

11 . The method according to claim 8 , wherein the reference image (I r ) and the updated image (I u ) are images taken at different times and/or at different frequencies and/or at different polarizations.

12 . The method according to claim 1 , wherein the image (I) is a radar image.

13 . The method according to claim 12 , wherein the probability model for magnitude assuming that a target is present is a reflectivity probability model.

14 . Method according to claim 1 , wherein the image (I) is a whole image (I) or a sub image of an image (I).

15 . Method according to claim 1 , wherein the image (I) is a Synthetic Aperture Radar, SAR, image (I).

16 . The method according to claim 15 , wherein the image (I) is obtained by synthetic aperture radar, SAR, operating below 500 MHz.

17 . A device arranged to detect at least one target in an image (I), wherein the image (I) comprises a set of pixels with an magnitude assigned to each pixel, wherein the device comprises processing circuitry ( 11 ) configured to, for a number of candidate targets K≥0, and starting with K=0, performing the following until the K+1th target does not show a probability (P) increase above a predetermined threshold value:
creating (S1) a candidate free image by removing a footprint of the candidate target K from the candidate free image of the previous candidate target K−1; wherein if K=0, the candidate free image is equal to the image (I);
calculating (S2), for the candidate free image, the probability of there being a target at each pixel, by using Bayes theorem; wherein Bayes theorem is calculated using the number of targets K;
determining (S3) a location of the candidate target K+1 in the image (I) by:
identifying (S31) the location in the candidate free image having a maximum probability of there being a target at each pixel, the identified location being the location of the candidate target K+1;

determining (S4), for each candidate target from 1 to K, the probability (P) that there is a target at the determined location associated with that candidate target by determining (S41) the calculated probability of there being a target at the determined location;
repeating the above with K substituted with K+1.

18 . A non-transitory computer readable medium storing a program, which, when executed on a device, causes the network node to perform the method according to claim 1 .

Assignees

• Saab Ab

Inventors

• Hellsten Hans

Classifications

• G06V10/764

  using classification, e.g. of video objects · CPC title

• G01S7/414

  Discriminating targets with respect to background clutter · CPC title

• G06F18/24155

  Bayesian classification · CPC title

• G06V20/13

  Satellite images · CPC title

• G06K9/6278

  Physics · mapped topic