Visibility determinations in physical space using evidential illumination values
Systems and methods are disclosed for identifying visibility of a physical space using visibility values and indications of uncertainty associated with the visibility values. One method can include capturing an image of a physical space, dividing pixels of the image into presumably well-lit and presumably not well-lit categories based on an intensity threshold, generating an evidential illumination map for the image based at least partly on a comparison between a value of each pixel to the intensity threshold, and projecting the evidential illumination map onto data representing a three-dimensional scan of the physical space. Evidential values can enable an autonomous vehicle to more safely navigate spaces using camera data by enabling programmatic determination of uncertainty for the camera data.
1. A computer-implemented method, as implemented by one or more processors, comprising:
receiving, by the one or more processors, first data corresponding to an image of a physical space captured by a camera;
calculating, by the one or more processors, an intensity threshold for the first data distinguishing well-lit pixels within the first data from not-well-lit pixels within the first data, wherein calculating the intensity threshold for the first data comprises converting the first data to greyscale, and wherein converting the first data to greyscale comprises generating a transmittance map for the first data;
generating, by the one or more processors, an evidential illumination map for the image based at least partly on a comparison between a value of each pixel in first data to an intensity threshold, wherein the evidential illumination map indicates an uncertainty that one or more pixels of the image depict a well-lit area;
obtaining, by the one or more processors, second data corresponding to a three-dimensional representation of the physical space;
using the second data, projecting, by the one or more processors, the evidential illumination map onto the physical space; and
identifying, by the one or more processors, a visibility of one or more positions in the physical space based on projection of the evidential illumination map onto the physical space.
2. The computer-implemented method of claim 1 , wherein projecting the evidential illumination map onto the physical space generates a grid associating positions within a coordinate system to visibility values and indications of uncertainty associated with the visibility values.
3. The computer-implemented method of claim 2 , wherein the indications of uncertainty associated with the visibility values are at least one of indications of evidence of being well lit, indications of evidence of being not well lit, or indications of uncertainty associated with a determination of being well-lit or not well-lit.
4. The computer-implemented method of claim 1 , wherein calculating the intensity threshold for the first data distinguishing well-lit pixels within the first data from not-well-lit pixels within the first data comprises applying an automatic thresholding algorithm to the first data.
5. The computer-implemented method of claim 4 , wherein the automatic thresholding algorithm is Otsu's method.
6. The computer-implemented method of claim 1 , wherein generating the evidential illumination map for the image is further based at least partly on a standard deviation of values for at least a subset of pixels within the first data.
7. The computer-implemented method of claim 1 , wherein converting the first data to greyscale comprises converting the first data to values within the L*a*b color space and generating a greyscale image based on L values within the lab color space.
8. The computer-implemented method of claim 1 , wherein generating a transmittance map for the first data comprises generating the transmittance map from at least one of a bright channel for the first data or a dark channel for the first data.
9. The computer-implemented method of claim 1 , wherein generating a transmittance map for the first data comprises generating the transmittance map based on passing the first data through a neural network machine learning model.
10. The computer-implemented method of claim 1 , wherein the second data is generated based on a lidar scan of the physical space.
11. A computing device comprising:
a data store storing computer-executable instructions; and
one or more processors configured to execute the computer-executable instructions, wherein execution of the computer-executable instructions causes the computing device to:
receive, by the one or more processors, first data corresponding to an image of a physical space captured by a camera;
calculate, by the one or more processors, an intensity threshold for the first data distinguishing well-lit pixels within the first data from not-well-lit pixels within the first data, wherein calculating the intensity threshold for the first data distinguishing well-lit pixels within the first data from not-well-lit pixels within the first data comprises applying an automatic thresholding algorithm to the first data. and wherein converting the first data to greyscale comprises generating a transmittance map for the first data;
generate, by the one or more processors, an evidential illumination map for the image based at least partly on a comparison between a value of each pixel in first data to an intensity threshold, wherein the evidential illumination map indicates an uncertainty that one or more pixels of the image depict a well-lit area;
obtain, by the one or more processors, second data corresponding to a three-dimensional representation of the physical space;
using the second data, project, by the one or more processors, the evidential illumination map onto the physical space; and
identify, by the one or more processors, a visibility of one or more positions in the physical space based on projection of the evidential illumination map onto the physical space.
12. The computing device of claim 11 , wherein projecting the evidential illumination map onto the physical space generates a grid associating positions within a coordinate system to visibility values and indications of uncertainty associated with the visibility values.
13. One or more non-transitory computer-readable media comprising computer-executable instructions that, when executed by a computing system comprising one or more processors, causes the computing system to:
receive, by the one or more processors, first data corresponding to an image of a physical space captured by a camera;
calculate, by the one or more processors, an intensity threshold for the first data distinguishing well-lit pixels within the first data from not-well-lit pixels within the first data, wherein calculating the intensity threshold for the first data distinguishing well-lit pixels within the first data from not-well-lit pixels within the first data comprises applying an automatic thresholding algorithm to the first data, and wherein calculating the intensity threshold for the first data comprises converting the first data to greyscale;
generate, by the one or more processors, an evidential illumination map for the image based at least partly on a comparison between a value of each pixel in first data to an intensity threshold, wherein the evidential illumination map indicates an uncertainty that one or more pixels of the image depict a well-lit area;
obtain, by the one or more processors, second data corresponding to a three-dimensional representation of the physical space;
using the second data, project, by the one or more processors, the evidential illumination map onto the physical space; and
identify, by the one or more processors, a visibility of one or more positions in the physical space based on projection of the evidential illumination map onto the physical space.
14. The one or more non-transitory computer-readable media of claim 13 , wherein projecting the evidential illumination map onto the physical space generates a grid associating positions within a coordinate system to visibility values and indications of uncertainty associated with the visibility values.