Automatic generation of vector map for vehicle navigation
A system will generate a vector map of a geographic area using a method that includes receiving a birds-eye view image of a geographic area. The birds-eye view image comprises various pixels. The system will process the birds-eye view image to generate a spatial graph representation of the geographic area, and it will save the node pixels and the lines to a vector map data set. The processor may be a component of a vehicle such as an autonomous vehicle. If so, the system may use the vector map data set to generate a trajectory for the vehicle as the vehicle moves in the geographic area.
1. A method of generating a vector map of a geographic area, the method comprising, by a processor:
acquiring a birds-eye view image of the geographic area, wherein the birds-eye view image comprises a plurality of pixels;
processing the birds-eye view image to generate a spatial graph representation of the geographic area by:
classifying each pixel of the birds-eye view image as either a lane pixel or a non-lane pixel,
processing the lane pixels to extract a plurality of features of the geographic area that are to be represented in the vector map,
for each of the lane pixels, determining a likelihood value that the lane pixel contains a feature of the geographic area that is to be represented in the vector map,
classifying the lane pixels having likelihood values that exceed a threshold as node pixels, and
generating the vector map by generating a plurality of lines such that each of the lines connects a corresponding pair of the node pixels, wherein each of the lines represents a lane segment in the geographic area, wherein generating the lines that connect the pairs of the node pixels comprises:
identifying, as the pairs of the node pixels, any pair of adjacent node pixels having features that indicate a relationship of spatial similarity; and
generating a matrix in which each element of the matrix has a value that indicates whether the pair of adjacent node pixels that the element represents are connected;
saving the vector map comprising nodes representing the node pixels and the lines to a vector map data set; and
moving an autonomous vehicle along a trajectory generated using the vector map data set.
2. The method of claim 1 , wherein the plurality of features of the geographic area comprise a lane border.
3. The method of claim 1 , further comprising: before generating the lines that connect the pairs of the node pixels, transforming each of the node pixels into a one-dimensional token.
4. The method of claim 3 , wherein each one-dimensional token is a vector representing one or more characteristics that are present at the geographic area at a spatial location that corresponds to the node pixel.
5. The method of claim 1 , wherein processing the lane pixels to extract the plurality of features of the geographic area is performed by employing a neural network.
6. A system for generating a vector map of a geographic area, the system comprising:
a processor: and
a memory device containing programming instructions that are configured to cause the processor to:
receive a birds-eye view image of the geographic area, wherein the birds-eye view image comprises a plurality of pixels;
process the birds-eye view image to generate a spatial graph representation of the geographic area by:
classifying each pixel of the birds-eye view image as either a lane pixel or a non-lane pixel,
processing the lane pixels to extract a plurality of features of the geographic area that are to be represented in the vector map,
for each of the lane pixels, determining a likelihood value that the lane pixel contains a feature of the geographic area that is to be represented in the vector map,
classifying the lane pixels having likelihood values that exceed a threshold as node pixels, and
generating the vector map by generating a plurality of lines such that each of the lines connects a corresponding pair of the node pixels, wherein each of the lines represents a lane segment in the geographic area, wherein generating the lines that connect the pairs of the node pixels comprises:
identifying, as the pairs of the node pixels, any pair of adjacent node pixels having features that indicate a relationship of spatial similarity; and
generating a matrix in which each element of the matrix has a value that indicates whether the pair of adjacent node pixels that the element represents are connected;
save the vector map comprising nodes representing the node pixels and the lines to a vector map data set; and
move a vehicle along a trajectory generated using the vector map data set.
7. The system of claim 6 , wherein the plurality of features of the geographic area comprise a lane border.
8. The system of claim 6 , wherein the memory device further contains additional programming instructions that are configured to cause the processor to, before generating the lines that connect the pairs of the node pixels, transform each of the node pixels into a one-dimensional token.
9. The system of claim 8 , wherein each one-dimensional token is a vector representing one or more characteristics that are present at the geographic area at a spatial location that corresponds to the node pixel.
10. The system of claim 6 , wherein processing the lane pixels to extract the plurality of features of the geographic area is performed by employing a neural network.
11. A computer program product for generating a vector map of a geographic area, the computer program product comprising a memory device containing programming instructions that are configured to cause a processor to:
receive a birds-eye view image of the geographic area, wherein the birds-eye view image comprises a plurality of pixels;
process the birds-eye view image to generate a spatial graph representation of the geographic area by:
classifying each pixel of the birds-eye view image as either a lane pixel or a non-lane pixel,
processing the lane pixels to extract a plurality of features of the geographic area that are to be represented in the vector map,
for each of the lane pixels, determining a likelihood value that the lane pixel contains a feature of the geographic area that is to be represented in the vector map,
classifying the lane pixels having likelihood values that exceed a threshold as node pixels, and
generating the vector map by generating a plurality of lines such that each of the lines connects a corresponding pair of the node pixels, wherein each of the lines represents a lane segment in the geographic area, wherein generating the lines that connect the pairs of the node pixels comprises:
identifying, as the pairs of the node pixels, any pair of adjacent node pixels having features that indicate a relationship of spatial similarity; and
generating a matrix in which each element of the matrix has a value that indicates whether the pair of adjacent node pixels that the element represents are connected;
save the vector map comprising nodes representing the node pixels and the lines to a vector map data set; and
move a vehicle along a trajectory generated using the vector map data set.
12. The computer program product of claim 11 , wherein the memory device further contains additional programming instructions that are configured to cause the processor to, before generating the lines that connect the pairs of the node pixels, transform each of the node pixels into a one-dimensional token, wherein each one-dimensional token is a vector representing one or more characteristics that are present at the geographic area at a spatial location that corresponds to the node pixel.