Photogrammetric Relief Modeling With Multi-Stereo Disparity Optimization

The patent merges the processing of multiple stereo image pairs by directly integrating their correlation results into a single relief representation model through iterative optimization, rather than converting each pair separately and then merging the models. This combines the information from all stereo pairs at the correlation level, preserving overhanging terrain structures and minimizing information loss while maintaining computational feasibility through the iterative gradient descent approach.

The patent performs the grid sampling and discretization as a preliminary action only at the final stage, after all correlation results from multiple stereo image pairs have been integrated through iterative optimization. The continuous relief model is then sampled onto a digital grid, which preserves information precision because the sampling occurs after the complete information integration, rather than requiring intermediate re-sampling operations that would degrade the data.

The patent implements iterative gradient descent optimization to minimize the cost function representing information loss, but applies partial action by using practical stopping criteria such as maximum iteration limits or convergence thresholds. This allows the optimization to improve model accuracy by integrating correlation results from multiple stereo pairs, while preventing excessive processing time by terminating iterations when sufficient accuracy is achieved or a predefined maximum is reached.