Automatic multi-image 3D ground control point extraction

A non-transitory machine-readable medium including instructions that, when executed by a machine, cause a machine to perform operations for determining a multi-image three-dimensional (3D) ground control point (GCP), the operations comprising:

• extracting, from a first two-dimensional (2D) image including image data of a first geographical region, a first image template, the first image template including a contiguous subset of pixels from the first 2D image and a first pixel of the contiguous subset of pixels of the first 2D image indicated by the GCP, wherein the GCP is a five-tuple that associates a point in a three-dimensional (3D) point set with the first pixel of the first image;
• predicting a first pixel location of the first pixel of the GCP in a second 2D image, the second 2D image including image data of a second geographical region overlapping with the first geographical region;
• extracting, from the second 2D image, a second image template, the second image template including a contiguous subset of pixels from the second 2D image and a second pixel of the contiguous subset of pixels of the second 2D image corresponding to the first pixel location;
• determining respective correlation scores, using a normalized cross-correlation, for the first image template centered at a variety of pixels of the contiguous subset of pixels of the second image template;
• comparing a highest correlation score of the correlation scores to a first threshold value;
• discarding a location of the second pixel responsive to determining a ratio is less than the first threshold value;
• identifying an undiscarded highest correlation pixel of the second 2D image corresponding to a remaining highest correlation score;
• and
• adding a pixel location of the undiscarded highest correlation pixel to the GCP making the GCP a seven-tuple to register the second 2D image to the 3D point set.

A system comprising:

• a memory including image data of first and second images of a geographical region stored thereon;
• and
• processing circuitry coupled to the memory, the processing circuitry configured to perform steps comprising:
• extracting, from a first two-dimensional (2D) image including image data of a first geographical region, a first image template, the first image template including a contiguous subset of pixels from the first 2D image and a first pixel of the contiguous subset of pixels of the first 2D image indicated by the GCP, wherein the GCP is a five-tuple that associates a point in the 3D point set with the first pixel of the first image;
• predicting a first pixel location of the first pixel of the GCP in a second 2D image, the second 2D image including image data of a second geographical region overlapping with the first geographical region;
• extracting, from the second 2D image, a second image template, the second image template including a contiguous subset of pixels from the second 2D image and a second pixel of the contiguous subset of pixels of the second 2D image corresponding to the first pixel location;
• determining respective correlation scores, using a normalized cross-correlation, for the first image template centered at a variety of pixels of the contiguous subset of pixels of the second image template;
• comparing a highest correlation score of the correlation scores to a first threshold value;
• discarding a location of the second pixel responsive to determining the ratio is less than the first threshold value;
• identifying an undiscarded highest correlation pixel of the second 2D image corresponding to a remaining highest correlation score;
• and
• adding a pixel location of the undiscarded highest correlation pixel to the GCP making the GCP a seven-tuple to register the second 2D image to the 3D point set.