Entrance detection from street-level imagery
View Patent ↗Architecture that detects entrances on building facades. In a first stage, scene geometry is exploited and the multi-dimensional problem is reduced down to a one-dimensional (1D) problem. Entrance hypotheses are generated by considering pairs of locations along lines exhibiting strong gradients in the transverse direction. In a second stage, a rich set of discriminative image features for entrances is explored according to constructed designs, specifically focusing on properties such as symmetry and color consistency, for example. Classifiers (e.g., random forest) are utilized to perform automatic feature selection and entrance classification. In another stage, a joint model is formulated in three dimensions (3D) for entrances on a given facade, which enables the exploitation of physical constraints between different entrances on the same facade in a systematic manner to prune false positives, and thereby select an optimum set of entrances on a given facade.
1. A computer-implemented method for detecting entrances in images of structural facades, the method comprising:
capturing images of a facade of a building;
capturing a LIDAR map of the facade;
generating a foreground mask for the captured images based on the LIDAR map;
applying the foreground mask to the captured images to generate masked images;
extracting one or more entrance candidates from the masked images by detecting vertical edgelets in the masked images using edge detection, a vertical edgelet including an edge pixel and one or more neighboring pixels along a vertical axis; and
selecting one or more true candidates from the entrance candidate based on known physical constraints of entrances.
2. The computer-implemented method of claim 1 , wherein generating a foreground mask for the captured images based on the LiDar map further comprises:
determining a frontal view of the façade using the LiDar map; and
generating the foreground mask, the foreground mask masking out portions of the captured images outside the frontal view of the façade.
3. The computer-implemented method of claim 1 , wherein generating a foreground mask for the captured images based on the LiDar map further comprises:
determining a ground line using the LiDar map; and
generating the foreground mask, the foreground mask masking out portions of the captured images below the ground line.
4. The computer-implemented method of claim 1 , wherein extracting one or more entrance candidates from the masked images further comprises:
separating entrances from non-entrances using a classifier, the classifier performing automatic feature selection and entrance classification to remove non-entrances.
5. The computer-implemented method of claim 4 , wherein features used in the classifier include degree of entrance symmetry and entrance color.
6. The computer-implemented method of claim 1 , wherein extracting one or more entrance candidates from the masked images further comprises:
generating a distribution of all detected vertical edgelets;
extracting a number of top ranked local peaks from the distribution to determine one or more vertical boundaries; and
determining one or more entrance candidates based on the one or more vertical boundaries.
7. The computer-implemented method of claim 6 , further comprising:
detecting horizontal edgelets in the masked images using edge detection;
generating a second distribution of all detected horizontal edgelets;
extracting a number of top ranked local peaks from the second distribution to determine one or more horizontal boundaries; and
selecting one or more entrance candidates, each entrance candidate having a pair of vertical boundaries within a predefined ratio of length to a horizontal boundary of the one or more horizontal boundaries.
8. The computer-implemented method of claim 1 , wherein selecting one or more true candidates from the entrance candidate further comprises:
projecting the entrance candidates into multi-dimensional space to resolve conflicts based on the known physical constraints using a multi-dimensional reasoning component, the known physical constraints encoded within the a multi-dimensional reasoning component.
9. A system for detecting entrances in images of structural facades, comprising:
at least one processor configured to execute computer-executable instructions in a computer readable hardware storage memory, the computer-executable instructions, when executed by the processor, cause the processor to:
capture images of a façade of a building;
capture a LIDAR map of the façade;
generate a foreground mask for the captured images based on the LIDAR map;
apply the foreground mask to the captured images to generate masked images;
extract one or more entrance candidates from the masked images by detecting vertical edgelets in the masked images using edge detection, a vertical edgelet including an edge pixel and one or more neighboring pixels along a vertical axis; and
select one or more true candidates from the entrance candidate based on known physical constraints of entrances.
10. The system of claim 9 , wherein the computer readable hardware storage memory comprises further computer-executable instructions for the generation of the foreground mask for the captured images based on the LiDar map, that when executed by the processor, cause the processor to:
determine a frontal view of the façade using the LiDar map; and
generate the foreground mask, the foreground mask masking out portions of the captured images outside the frontal view of the façade.
11. The system of claim 9 , wherein the computer readable hardware storage memory comprises further computer-executable instructions for the generation of the foreground mask for the captured images based on the LiDar map, that when executed by the processor, cause the processor to:
determine a ground line using the LiDar map; and
generate the foreground mask, the foreground mask masking out portions of the captured images below the ground line.
12. The system of claim 9 , wherein the computer readable hardware storage memory comprises further computer-executable instructions for the extraction of the one or more entrance candidates from the masked images, that when executed by the processor, cause the processor to:
separate entrances from non-entrances using a classifier, the classifier performing automatic feature selection and entrance classification to remove non-entrances.
13. The system of claim 12 , wherein features used in the classifier include degree of entrance symmetry and entrance color.
14. The system of claim 9 , wherein the computer readable hardware storage memory comprises further computer-executable instructions for the extraction of the one or more entrance candidates from the masked images, that when executed by the processor, cause the processor to:
generate a distribution of all detected vertical edgelets;
extract a number of top ranked local peaks from the distribution to determine one or more vertical boundaries; and
determine one or more entrance candidates based on the one or more vertical boundaries.
15. The system of claim 14 , wherein the computer readable hardware storage memory comprises further computer-executable instructions, that when executed by the processor, cause the processor to:
detect horizontal edgelets in the masked images using edge detection;
generate a second distribution of all detected horizontal edgelets;
extract a number of top ranked local peaks from the second distribution to determine one or more horizontal boundaries; and
select one or more entrance candidates, each entrance candidate having a pair of vertical boundaries within a predefined ratio of length to a horizontal boundary of the one or more horizontal boundaries.
16. A computer-readable hardware storage medium comprising computer-executable instructions for detecting entrances in images of structural facades that when executed by a processor, cause the processor to:
capture images of a façade of a building;
capture a LIDAR map of the façade;
generate a foreground mask for the captured images based on the LIDAR map;
apply the foreground mask to the captured images to generate masked images;
extract one or more entrance candidates from the masked images by detecting vertical edgelets in the masked images using edge detection, a vertical edgelet including an edge pixel and one or more neighboring pixels along a vertical axis; and
select one or more true candidates from the entrance candidate based on known physical constraints of entrances.
17. The computer-readable hardware storage medium of claim 16 , comprising further computer-executable instructions for the generation of the foreground mask for the captured images based on the LiDar map, that when executed by the processor, cause the processor to:
determine a frontal view of the façade using the LiDar map; and
generate the foreground mask, the foreground mask masking out portions of the captured images outside the frontal view of the façade.
18. The computer-readable hardware storage medium of claim 16 , comprising further computer-executable instructions for the generation of the foreground mask for the captured images based on the LiDar map, that when executed by the processor, cause the processor to:
determine a ground line using the LiDar map; and
generate the foreground mask, the foreground mask masking out portions of the captured images below the ground line.
19. The computer-readable hardware storage medium of claim 16 , comprising further computer-executable instructions for the extraction of the one or more entrance candidates from the masked images, that when executed by the processor, cause the processor to:
separate entrances from non-entrances using a classifier, the classifier performing automatic feature selection and entrance classification to remove non-entrances.
20. The computer-readable hardware storage medium of claim 16 , comprising further computer-executable instructions for the extraction of the one or more entrance candidates from the masked images, that when executed by the processor, cause the processor to:
generate a distribution of all detected vertical edgelets;
extract a number of top ranked local peaks from the distribution to determine one or more vertical boundaries; and
determine one or more entrance candidates based on the one or more vertical boundaries.