Mapping pipeline optimization using aggregated overhead view reconstruction

1. A method comprising:

obtaining an aggregated image of a geographical area that has been generated by a pipeline as an overhead view wherein the aggregated image is generated from aggregating pixel values from source images of the geographical area including a plurality of regions;

generating, from the aggregated image, one or more reprojection images of one or more of the regions;

calculating a difference between the source images of the geographical area and the one or more reprojection images; and

determining, according to the difference, one or more error corrections to be applied to the pipeline for generating overhead view images, including:

performing an iterative optimization process to identify correction parameters that minimize a reprojection error when utilized by the pipeline;

identifying a parameter difference between original parameters used by the pipeline and the correction parameters; and

using the parameter difference to derive the error corrections.

2. The method as recited in claim 1 , wherein the source images from which the aggregated image is generated comprises one or more source images captured at a substantially ground-level view.

3. The method as recited in claim 1 , wherein the one or more reprojection images are generated at a substantially ground-level view.

4. The method as recited in claim 1 , wherein calculating one or more differences comprises an analysis of one or more of: color space and a distance function.

5. The method as recited in claim 1 , wherein the error corrections comprise one or more modifications to one or more parameters utilized by the pipeline to generate one or more of the aggregated image.

6. The method as recited in claim 1 , wherein the error corrections comprise a modification to (i) one or more parameters related to capture of image data utilized by the pipeline to generate aggregated overhead view images, (ii) one or more parameters related to capture of point cloud data utilized by the pipeline to generate aggregated overhead view images, (iii) one or more parameters related to calibration or synchronization of sensor units utilized by the pipeline to generate aggregated overhead view images, (iv) one or more parameters related to the generation of simultaneous localization and mapping (SLAM) information utilized by the pipeline to generate aggregated overhead view images, and (v) one or more parameters related to one or more processing techniques utilized by the pipeline to generate aggregated overhead view images.

7. The method as recited in claim 6 , wherein the one or more processing techniques utilized by the pipeline to generate the aggregated image comprises a segmentation technique.

8. The method as recited in claim 1 , wherein identifying the source images comprise one or more of: (i) filtering out one or more of the source images taken at a distance greater than a predetermined distance; (ii) filtering out a portion of one or more of the source images including over a predetermined amount of shadow; (iii) filtering out one or more of the source images below a predetermined quality threshold; (iv) filtering out one or more of the source images taken within a predetermined time window; (v) filtering out one or more of the source images taken from a given camera; and (vi) filtering out pixels that are masked out by one or more segmentation masks.

9. The method as recited in claim 8 , wherein the one or more segmentation masks comprises a mask corresponding to one or both of a road surface or a vehicle mask.

10. The method as recited in claim 1 , wherein calculating the difference between the source images and the reprojection images comprises:

summing difference values per pixel; and

outputting one or more error metrics.

11. The method as recited in claim 10 , wherein calculating the difference includes considering pixels in a predetermined limit around edges of the source images, and wherein the predetermined limit comprises a predetermined size of a pixel grid.

12. The method as recited in claim 1 , further comprising:

determining an image quality metric per pixel with respect to the aggregated image.

13. The method as recited in claim 12 , wherein generating the one or more reprojection images occurs according to the image quality metric.

14. A non-transitory computer-readable medium comprising program instructions stored thereon that are executable by at least one processor such that a computing system is capable of:

obtaining an aggregated overhead view image of a geographical area that has been generated by a pipeline, the geographical area comprising a plurality of regions, wherein the aggregated overhead view image is generated from aggregating pixel values from a plurality of source images of the geographical area;

generating one or more reprojection images of one or more of the regions of the geographic area from the aggregated overhead view image;

calculating a difference between the source images of the geographical area and the one or more reprojection images; and

determining, according to the difference, error corrections to be applied to the pipeline for generating overhead view images, including:

performing an iterative optimization process to identify correction parameters that minimize a reprojection error when utilized by the pipeline;

identifying a parameter difference between original parameters used by the pipeline and the correction parameters; and

using the parameter difference to derive the error corrections.

15. The computer-readable medium of claim 14 , wherein the one or more error corrections comprise a modification to (i) one or more parameters related to capture of image data utilized by the pipeline to generate aggregated overhead view images, (ii) one or more parameters related to capture of point cloud data utilized by the pipeline to generate aggregated overhead view images, (iii) one or more parameters related to calibration or synchronization of sensor units utilized by the pipeline to generate aggregated overhead view images, (iv) one or more parameters related to the generation of simultaneous localization and mapping (SLAM) information utilized by the pipeline to generate aggregated overhead view images, (v) one or more parameters related to one or more processing techniques utilized by the pipeline to generate aggregated overhead view images.

16. The computer-readable medium of claim 14 , wherein the one or more processing techniques utilized by the pipeline to generate aggregated overhead view images comprises a segmentation technique.

17. A computing system comprising:

at least one processor;

a non-transitory computer-readable medium; and

program instructions stored on the non-transitory computer-readable medium that are executable by the at least one processor such that the computing system is capable of:

obtaining an aggregated overhead view image of a geographical area that has been generated by a pipeline, the geographical area comprising a plurality of regions, wherein the aggregated overhead view image is generated from aggregating pixel values from a plurality of source images of the geographical area;

generating one or more reprojection images of one or more of the regions of the geographic area from the aggregated overhead view image;

calculating a difference between the source images and the one or more reprojection images; and

determining, according to the difference, one or more error corrections to be applied to the pipeline for generating overhead view images, including:

performing an iterative optimization process to identify correction parameters that minimize a reprojection error when utilized by the pipeline;

identifying a parameter difference between original parameters used by the pipeline and the correction parameters; and

using the parameter difference to derive the error corrections.