DETECTING AND IDENTIFYING PARKING LOTS IN REMOTELY-SENSED IMAGES

• US 20150104070A1
• Filed: 10/14/2013
• Published: 04/16/2015
• Est. Priority Date: 10/14/2013
• Status: Active Grant

• Current Assignee: Maxar Intelligence Incorporated (Galileo Parent, Inc.)
• Original Assignee: DigitalGlobe, Inc. (Galileo Parent, Inc.)
• Inventors: Tabb, Mark
• Granted Patent: US 9,275,297 B2
• Time in Patent Office: Days
• US Class Current: 382/104
• CPC Class Codes: G06V 10/48 by mapping characteristic values of the pattern into a parameter space, e.g. Hough transformation
• CPC Class Codes: G06V 20/176 Urban or other man-made structures

Abstract

Processing and analyzing at least one remotely-sensed image to automatically detect and identify parking lots within a region of the remotely-sensed image. This may include: (1) receiving a remotely-sensed image having zero or more parking lots; (2) identifying a set of pixels related to parking lot features within the remotely-sensed image; (3) identifying at least one parking row within the remotely-sensed image; and (3) identifying a parking lot based on the identified parking rows. Identifying at least one parking lot may further include using first, second, third, and fourth level modules, such as a concrete/asphalt detection module, a road marking detection module, a vehicle detection module, and a Hough aggregation for parking row detection module.

Claims

1. A method for identifying parking lots in a remotely-sensed image, the method comprising:

receiving at least one remotely-sensed image having zero or more parking lots;

identifying a first set of pixels related to a parking lot surface material within the remotely-sensed image with a first level module;

identifying a second set of pixels related to parking lot lines within the remotely-sensed image with a second level module;

identifying at least one parking lot object of interest within the remotely-sensed image with a third level module;

identifying at least one parking row within the remotely-sensed image with a fourth level module, wherein the fourth level module detects the at least one parking row based on data received from the first, second, and third level modules; and

identifying a parking lot based on information from the first, second, third, and fourth modules.

2. The method according to claim 1, further comprising compensating for materials having spatial and temporal variability between remotely-sensed images having different domains.

3. The method according to claim 1, wherein identifying the first set of pixels with the first level module comprises:

classifying the first set pixels with a decision tree classifier.

4. The method according to claim 1, wherein identifying the second set of pixels with the second level module comprises:

identifying an area of subpixels that are locally brighter than surrounding subpixels; and

checking that the subpixels are not exceeding a resolution of one pixel.

5. The method according to claim 1, wherein identifying the at least one parking lot object with the third level module comprises:

developing a first set of weak classifiers based on Histogram of Oriented Gradient features;

developing a second set of weak classifiers based on Haar features;

training the first and second set of weak classifiers to detect the at least one parking lot object of interest; and

cascading the first and second set of weak classifiers to improve detection accuracy.

6. The method according to claim 1, wherein identifying the at least one parking row with the fourth level module comprises:

establishing a spatial regularity property related to parking rows;

assigning at least one parameter related to the parking rows; and

detecting a parking row based on the established spatial regularity property and the at least one parameter.

7. The method according to claim 1 further comprising providing parking lot parameters.

8. The method according to claim 1, wherein the fourth level module identifies parking rows with an accuracy rate of at least 90%.

9. The method according to claim 1, wherein the first level module can achieve an accuracy rate of at least 90%.

10. The method according to claim 1, wherein the second level module is applied to a panchromatic image.

11. The method according to claim 1, wherein the second level module is applied to a short-wave infrared image.

12. The method according to claim 1, wherein the fourth level module assigns the at least one parameter related to the parking rows by analyzing a 2-D position, a row length, a first row orientation, a second row orientation angle corresponding to the orientation of the parking lot versus the row, a spacing representing the average width of a parking lot, and a parking lot stripe spacing representing the distance between the parking lot stripes on adjacent parking lots.

13. A method for identifying parking rows in a remotely-sensed image, the method comprising:

receiving at least one remotely-sensed image having zero or more parking lots;

exploiting at least one spatial regularity property commonly found in parking rows;

analyzing the at least one remotely-sensed image based on the exploited spatial regularity property; and

identifying zero or more parking rows in the at least one remotely-sensed image;

14. The method according to claim 13, further comprising compensating for a bidirectional reflectance distribution between remotely-sensed images having different domains.

15. The method according to claim 13, further comprising identifying a first set of pixels related to a parking lot surface material with a concrete/asphalt detection module.

16. The method according to claim 15, wherein identifying the first set of pixels related to the parking lot surface material comprises:

classifying the first set pixels with a decision tree classifier.

17. The method according to claim 13, further comprising identifying a second set of pixels related to marking lines with a road marking detection module.

18. The method according to claim 17, wherein identifying the second set of pixels with the road marking detection module comprises:

identifying an area of subpixels that are locally brighter than surrounding subpixels; and

checking that the subpixels are not exceeding a resolution of one pixel.

19. A method for identifying parking lots in a remotely-sensed image, the method comprising:

receiving a first remotely-sensed image having zero or more parking lots;

identifying a first set of pixels related to a parking lot surface material within the first remotely-sensed image with a first level module to create a first tagged image;

receiving a second remotely-sensed image having zero or more parking lots;

identifying a second set of pixels related to parking lot lines within the second remotely-sensed image with a second level module to create a second tagged image;

masking the first tagged image with the second tagged image to create a first masked image;

identifying at least one parking lot object of interest within the second remotely-sensed image with a third level module to create a third tagged image;

masking the first tagged image with the third tagged image to create a second masked image;

identifying at least one parking row within the first and second masked images with a fourth level module; and

identifying a parking lot based on the identified at least one parking row.

20. The method according to claim 19, further comprising compensating for materials having spatial and temporal variability between remotely-sensed images having different domains.