Case Study: Set Reconstruction

Project Target & Shot Conditions

The primary objective of this project was to reconstruct precise 3D environment geometries (geometry hulls) and solve the exact physical camera motion for several indoor and outdoor shots. The source footage was captured using a DSLR camera under varied, uncontrolled conditions without pre-recorded camera logs, focal length statistics, or lens calibration checkboard patterns.

Without on-set telemetry, the matchmove solver was tasked with calculating camera focal lengths, sensor dimensions, and distortion coefficients simultaneously with the 3D translation path. This required highly consistent tracking points across extreme depth planes to avoid mathematical ambiguity.

Adopted Workflow

1. Footage Preparation: Linearizing dynamic ranges to pull contrast in under-exposed shadows and over-exposed skies, maximizing trackable details.
2. Optical Flow Grid: Running an automated optical flow track to collect high-density background details, combined with manual tracking anchors on sharp static structures (doorframes, steps, wall corners).
3. Two-Stage Focal Solve: Solving the static camera orientation first to estimate the constant focal length, then solving the camera path coordinates dynamically.
4. 3D Set Reconstruction: Exporting the solved 3D point cloud into modeling environments to fit structural polygonal geometry (floors, columns, walls) matching the physical spaces.
5. Telemetry Validation: Re-projecting the low-poly models onto the footage to inspect alignment, verify lack of slippage, and validate reprojection error limits.

Solver Challenges & Solutions

Textureless Indoor Surfaces: White hallways presented a classic tracking challenge due to the lack of high-contrast natural features. To solve this, manual tracking points were locked onto structural lines, such as door thresholds, tile grout lines, and electrical fixtures.

Dynamic Outdoor Parallax: Outdoor courtyard shots featured dynamic depth shifts from foreground arches to distant trees. Parallax planes had to be carefully separated: foreground objects were isolated using rotoscoped masks to ensure the background camera solver wasn't corrupted by moving elements.

Camera Lens Distortion: DSLR kit lenses exhibit slight radial barrel distortion. The solver calculated a custom radial distortion model (`K1` and `K2` coefficients) to rectify the plates, ensuring the modeled straight lines aligned perfectly with the curved edges of the lens.