Nonlinear self-calibration for structure from motion (SFM) techniques

A system, comprising:

• one or more processors;
• and
• a memory comprising program instructions, wherein the program instructions are executable by at least one of the one or more processors to perform nonlinear self-calibration on a plurality of images in an image sequence to convert a projective reconstruction for the image sequence to a metric reconstruction of camera motion and camera intrinsic parameters for the image sequence, wherein, to perform nonlinear self-calibration, the program instructions are executable by at least one of the one or more processors to:
• determine at least two sets of initial values for an equation to be optimized according to a nonlinear optimization technique;
• optimize the equation using each set of initial values according to the nonlinear optimization technique to generate a result for each set of initial values, wherein each result includes estimates of one or more camera intrinsic parameters and of camera motion parameters;
• select the result with a smaller cost according to a cost function;
• and
• generate a metric reconstruction for each of the plurality of images according to the selected result, wherein the metric reconstruction includes an estimate of one or more camera intrinsic parameters and of camera motion parameters for each image.

A non-transitory computer-readable storage medium storing program instructions, wherein the program instructions are computer-executable to implement:

• performing nonlinear self-calibration on a plurality of images in an image sequence to convert a projective reconstruction for the image sequence to a metric reconstruction of camera motion and camera intrinsic parameters for the image sequence, wherein, in said performing nonlinear self-calibration, the program instructions are computer-executable to implement:
• determining at least two sets of initial values for an equation to be optimized according to a nonlinear optimization technique;
• optimizing the equation using each set of initial values according to the nonlinear optimization technique to generate a result for each set of initial values, wherein each result includes estimates of one or more camera intrinsic parameters and of camera motion parameters;
• selecting the result with a smaller cost according to a cost function;
• and
• generating a metric reconstruction for each of the plurality of images according to the selected result, wherein the metric reconstruction includes an estimate of one or more camera intrinsic parameters and of camera motion parameters for each image.