Method and system for dismount detection in low-resolution UAV imagery

Patent metadata

Field	Value
Publication number	US-10127448-B2
Application number	US-201514829171-A
Country	US
Kind code	B2
Filing date	Aug 18, 2015
Priority date	Aug 27, 2014
Publication date	Nov 13, 2018
Grant date	Nov 13, 2018

Abstract

A method for dismount detection in low-resolution UAV imagery, comprising providing an input image, processing a greyscale distribution of the input image, determining a rough classification in the input image based on the grayscale distribution, determining the optimal parameters based on the rough classification, estimating one or more potential dismount locations, applying an area filter to the one or more potential dismount locations, removing undesired locations from the one or more potential dismount locations, applying one or more secondary filters to the resulting one or more potential dismount locations, assigning a probability to the one or more potential dismount locations, and assessing desirability of the one or more potential dismount locations.

First claim

What is claimed is:

1. A method for dismount detection in a single low-resolution, high-altitude image, comprising;
providing a single input image;
processing a grayscale distribution of the single input image;
determining a rough classification of the single input image based on the grayscale distribution and without the use of any reference images;
determining optimal parameters based on the determined rough classification for the single input image without the use of any reference images;
estimating at least one potential dismount location;
extracting blobs from the single input image without the use of any reference images;
applying an area filter to the at least one potential dismount location;
removing undesired locations from the at least one potential dismount location;
applying at least one secondary filter to the resulting at least one potential dismount location wherein the at least one secondary filter comprises applying at least two of: a contrast filter, a dynamic filter, a thermal filter, a change detection filter, a motion history filter, and a clutter identification filter;
fusing the results of the at least two secondary filters;
assigning a probability to the at least one potential dismount location; and
assessing desirability for initiating tracking of the at least one potential dismount location.

2. The method of claim 1 , wherein the step of determining a rough classification further comprises classifying the single input image by a time of day without the use of any reference images.

3. The method of claim 1 , wherein the step of determining a rough classification, comprises classifying the single input image by a weather condition without the use of any reference images.

4. The method of claim 1 , wherein the single input image is generated from a camera carried on an aerial vehicle.

5. The method of claim 1 , wherein the step of determining the optimal parameters further comprises selecting at least one of the parameters from the group consisting of: image smoothing, contrast enhancement, and multi-level image segmentation.

6. A method to improve the reliability of dismount detection from a single, low resolution image, the method comprising the steps of:
obtaining a single input image; and,
determining from the single input image a probability that a portion of the single input image corresponds to a dismount by applying a plurality of filtering techniques and image processing techniques without the use of any reference images, and
fusing the results of the plurality of filtering techniques and image processing techniques to ascertain a probability of a dismount being present in the single input image without use of any reference images,
wherein the plurality of filtering techniques comprise at least one secondary filter, wherein the at least one secondary filter comprises at least one of a contrast filter, a dynamic filter, a thermal filter, a change detection filter, a motion history filter, and a clutter identification filter.

7. The method of claim 6 , wherein at least one of the plurality of filtering techniques comprises determining a rough classification of the single input image by type of scene illumination without the use of any reference images.

8. The method of claim 7 , further comprising using the rough classification to determine optimal filter parameters for the plurality of filtering techniques without the use of any reference images.

9. The method of claim 8 , wherein the optimal parameters include at least one of: image smoothing, contrast enhancement, and multilevel image segmentation.

10. The method of claim 6 , wherein the step of determining a probability further comprises extracting a circumscribed fuzzy area from the single input image without the use of any reference images and utilizing the circumscribed fuzzy area in at least one of the plurality of filtering techniques.

11. The method of claim 10 , wherein the circumscribed fuzzy area includes a blob, the at least one of the plurality of filtering techniques includes area filtering, and further comprising the step of applying area filtering to the blob, whereby portions of the single input image outside of the area determined by the area filtering are eliminated from evaluation without the use of any reference images.

12. The method of claim 11 , further comprising a shape filter applied to an output of the at least one secondary filter to establish whether a shape of the blob corresponds to a shape of a human being.

13. The method of claim 12 , further comprising the step of applying the shape filter only if the probability of the dismount is above a minimum probability and below a maximum probability.

14. The method of claim 6 , further comprising the step of adding a dismount to a potential dismount list when the probability of a dismount exceeds at least one threshold value.

15. The method of claim 14 , further including the step of ascertaining a location of a dismount added to the potential dismount list and utilizing the location of the dismount for dismount tracking.

16. The method of claim 7 , wherein the step of determining a rough classification further comprises image classification by first analyzing the single input image to determine whether the scene involves one of: a sunny day, a cast shadow day, and night imagery without the use of any reference images.

17. The method of claim 16 , wherein the image classification is used to dynamically adjust filter parameters prior to image smoothing, contrast enhancement, and multilevel segmentation.

18. The method of claim 17 , further comprising blob extraction after performing at least one of image smoothing, contrast enhancement and multilevel segmentation.

19. The method of claim 18 , further comprising filtering out extracted blobs that are outside a predetermined area.

20. The method of claim 19 , further comprising filtering the blobs that are within the predetermined area utilizing at least one of the secondary filters.

21. The method of claim 20 , wherein, by using the at least one secondary filters, a reliability of dismount detection is at least 98%.

22. The method of claim 21 , wherein the reliability of the dismount detection increases when the probability of a dismount passed through the at least one secondary filters exceeds a predetermined minimum.

23. The method of claim 22 , wherein the reliability of dismount detection increases if the probability of a dismount as determined by the at least one secondary filters exceeds a minimum and maximum probability.

24. The method of claim 21 , wherein the reliability of dismount detection is increased by performing a shape filtering function on outputs from the at least one secondary filters when the probability of a dismount as determined by the at least one secondary filters exceeds a minimum probability but does not exceed a maximum probability.

25. A method of improving the reliability of dismount detection in a single, low resolution input image, the method comprising the steps of:
receiving a single low resolution input image from a high altitude, wherein the single low resolution input image has a potential dismount;
processing the single low resolution input image utilizing rough classification filtering without the reference images;
processing the single low resolution input image without the use of any reference images using at least two of: image smoothing, contrast enhanceme

Assignees

• Bae Sys Inf & Elect Sys Integ

Inventors

• Hammoud Riad

Classifications

• G06F18/241

  relating to the classification model, e.g. parametric or non-parametric approaches · CPC title

• G06V20/13 Primary

  Satellite images · CPC title

• G06K9/0063 Primary

  Physics · mapped topic

• G06K9/6268

  Physics · mapped topic