Generation of high resolution population density data sets through exploitation of high resolution overhead imagery data and low resolution population density data sets
View Patent ↗Utilities (e.g., systems, methods, etc.) for automatically generating high resolution population density estimation data sets through manipulation of low resolution population density estimation data sets with high resolution overhead imagery data (e.g., such as overhead imagery data acquired by satellites, aircrafts, etc. of celestial bodies). Stated differently, the present utilities make use of high resolution overhead imagery data to determine how to distribute the population density of a large, low resolution cell (e.g., 1000 m) among a plurality of smaller, high resolution cells (e.g., 100 m) within the larger cell.
1. A method for use in generating population density data sets of a geographic region from overhead imagery data, comprising:
obtaining a plurality of multi-scale morphological image descriptors for each of a respective plurality of pixels of at least one input overhead image of a geographic region;
training, with a processor, a linear model with a) a plurality of population density estimates of a respective plurality of first portions of the geographic region and b) the multi-scale morphological image descriptors within each of the plurality of first portions to obtain a plurality of weighting values; and
determining, with the processor, a plurality of population density estimates of a respective plurality of second portions of the geographic region using a) the plurality of weighting values and b) the multi-scale morphological image descriptors within each of the plurality of second portions, wherein the second portions are smaller than the first portions.
2. The method of claim 1 , wherein the at least one input overhead image of the geographic region comprises at least one first input overhead image at a first spatial resolution, and wherein the obtaining step includes:
obtaining structures of interest from at least one second overhead image of the geographic area at a second spatial resolution less than the first spatial resolution;
generating a mask with the obtained structures of interest; and
applying the mask over the plurality of multi-scale morphological image descriptors to obtain multi-scale morphological image descriptors that identify the structures of interest, wherein the multi-scale morphological image descriptors that identify the structures of interest are utilized during the training step.
3. The method of claim 1 , wherein the training step includes:
decomposing the multi-scale morphological image descriptors within each of the first portions into a respective plurality of first vectors that each include a plurality of entries, wherein each entry includes a value that represents a frequency of one of a plurality of fixed values of the multi-scale morphological image descriptors within a respective one of the first portions; and
combining the plurality of first vectors into a matrix.
4. The method of claim 3 , wherein the training step includes:
generating a second vector that includes a plurality of entries, wherein each entry includes a value that represents one of the population density estimates of a respective one of the plurality of first portions of the geographic region.
5. The method of claim 4 , wherein the linear model comprises a difference between a) the second vector and b) a product of the matrix and a third vector that includes a plurality of entries, and wherein the training step includes:
determining a particular plurality of values of the respective plurality of entries of the third vector that minimizes the difference, wherein the particular plurality of values is the plurality of weighting values.
6. The method of claim 5 , wherein a number of the plurality of entries of the second vector is the same as a number of the plurality of entries of the third vector.
7. The method of claim 5 , wherein the plurality of weighting values are arranged in a weighting vector, and wherein the step of determining the plurality of population density estimates of the respective plurality of second portions includes, for each second portion:
decomposing the multi-scale morphological image descriptors within the second portion into a fourth vector that includes a plurality of entries, wherein each entry includes a value that represents a frequency of one of the plurality of fixed values of the multi-scale morphological image descriptors within the second portion;
manipulating the fourth vector with the weighting vector to obtain the population density estimate for the second portion.
8. The method of claim 7 , wherein the manipulating step includes:
obtaining an inner product of the fourth vector and the weighting vector; and
multiplying the inner product by a total quantity of pixels in the second portion of the at least one overhead image to obtain the population density estimate for the second portion.
9. The method of claim 1 , further comprising:
mapping the plurality of population density estimates of the respective plurality of second portions into a resultant image of the geographic region.
10. The method of claim 1 , further comprising for each first portion of the geographic region:
using the population density estimate of the first portion to constrain the population density estimates of the second portions of the first portion.
11. The method of claim 10 , wherein the using step includes:
normalizing the population density estimates of the second portions into a plurality of normalized population density estimates;
multiplying the population density estimate of the first portion by the normalized population density estimates to obtain corrected population density estimates for the second portions.
12. The method claim 1 , wherein the obtaining step includes:
decomposing, using the processor, the plurality of pixels of the at least one input overhead image into a plurality of connected components of a hierarchical data structure;
generating a plurality of differential area profile (DAP) vectors for the respective plurality of pixels of the at least one overhead image using the components of the hierarchical data structure, wherein each DAP vector includes a plurality of entries corresponding to intensity differences between the connected components in which the respective pixel is resident; and
using the DAP vector for each pixel to obtain at least one multi-scale morphological image descriptor for the pixel.
13. The method of claim 12 , wherein the hierarchical data structure comprises a plurality of levels of connected components, and wherein the connected components at each level include pixels having common intensities.
14. A system for generating high resolution population density estimation cells of a geographic region from low resolution population density estimation cells and high resolution overhead imagery of the geographic region, comprising:
a training engine, executable by a processor, that trains a linear model to obtain a plurality of weighting factors to be used for determining population density estimates of a plurality of high resolution cells within each of a plurality of low resolution cells of a geographic area, wherein the linear model includes a) a respective plurality of population density estimates of the plurality of low resolution cells and b) a plurality of multi-scale morphological image descriptors of a respective plurality of pixels of at least one input overhead image of the geographic region within each of the plurality of low resolution cells; and
an estimation engine, executable by the processor, that estimates a plurality of population density estimates of the respective plurality of high resolution cells within the low resolution cells of the geographic region using a) the plurality of weighting factors and b) a plurality of multi-scale morphological image descriptors within each of the plurality of high resolution cells.
15. The system of claim 14 , wherein the training engine decomposes the multi-scale morphological image descriptors within each of the low resolution cells into a respective plurality of first vectors that each include a plurality of entries, wherein each entry includes a value that represents a frequency of one of a plurality of fixed values of the multi-scale morphological image descriptors within a respective one of the low resolution cells, and wherein the training engine combines the plurality of first vectors into a matrix.
16. The system of claim 15 , wherein the training engine generates a second vector that includes a plurality of entries, wherein each entry includes a value that represents one of the population density estimates of a respective one of the plurality of low resolution cells of the geographic region.
17. The system of claim 16 , wherein the linear model comprises a difference between a) the second vector and b) a product of the matrix and a third vector that includes a plurality of entries, wherein the training engine determines a particular plurality of values of the respective plurality of entries of the third vector that minimizes the difference, wherein the particular plurality of values is the weighting factor.
18. The system of claim 17 , wherein the plurality of weighting factors are arranged in a weighting vector, and wherein the estimation engine estimates the plurality of population density estimates of the respective plurality of high resolution cells by:
decomposing the multi-scale morphological image descriptors within the high resolution cell into a fourth vector that includes a plurality of entries, wherein each entry includes a value that represents a frequency of one of the plurality of fixed values of the multi-scale morphological image descriptors within the second portion;
obtaining an inner product of the fourth vector and the weighting vector; and
multiplying the inner product by a total quantity of pixels in the high resolution cell of the at least one overhead image to obtain the population density estimate for the high resolution cell.