This section provides a brief outline of the procedure to follow for PhotoModeler Inverse Camera projects. The previous sections go into more detail and background. This section is a quick set of steps to go through when starting any forensic, accident reconstruction, or historic photo project.
Steps
• Review the photographs available
• How many photos?
• What angles between photos?
• How much overlap (shared points)?
• What is photo quality and sharpness?
• The critical dimensions, do they appear on multiple photos or on a known plane in a single photo?
• Review what is known
• 2D or 3D control points available?
• Survey data available or a survey can be done?
• Can 3D control be made from other knowns?
• Is there enough control to do project and have accuracy comparison? If using Control Points, do you have at least four Control Points on each unknown photo?
• Does the control cover the photos well (i.e., not all in small portion of photo)?
• If no control points, do you have strong perspective and can use Axes Constraints on parallel edges and lines?
• If a crush project, is car make/model/year known?
• Is there access to an exemplar vehicle?
• What is known about the camera that took the photos?
• Make a decision based on 1) and 2) if you can proceed based on number of photos, what is needed for the case and the amount and type of knowns
• Decide on the approach
• Single photo resection with measurement/mapping on plane?
• Multiple photo - inverse cameras and normal projects
• Is an exemplar needed? Front crush, side crush, using offsets.
• Combined crush/exemplar project or separate projects?
• How will accuracy of the result be determined? Are there enough left over unknowns? Can you do some logical consistency checks?
• If using an exemplar, do exemplar PhotoModeler project
• Do standard multiple photo, known camera, PhotoModeler project to map out points that will be key in doing InverseCamera on unknown photos, and to map out points that will help measure the crush.
• InverseCamera project: Start with single best photo
• Photo with most control visible (with control visible over depth)
• Photo that is sharpest
• Photo that covers most of the area of interest
• If using Control Points, Import the Control
• Lock control points down with negative precision or check box
• If exemplar project done with PhotoModeler, export a DXF or text file from PhotoModeler and re-import in the Control Import Dialog (same or different project). Also see the Crush tutorial for an example.
• Decide on which camera parameters to solve for
• With film and scanned prints you can usually determine format size before hand (at least to a good enough approximation).
• Start off solving for focal length only (then add principal point if greater precision is needed).
• Test the single photo solution if possible.
• Test using any known dimensions on planes.
• Review camera parameters to see if they make sense.
• Study the 3D viewer to see if it makes sense.
• Study marking residuals carefully. Esp. look for residuals changing in size as distance of point to camera changes (indicating a poor focal length solution).
• Add new photos and base new cameras on solved one
• Using the camera parameters from the first good solution make up new cameras for each photo if you think a zoom lens was involved.
• Do inverse camera on each new photo/camera combination allowing only focal length to be solved in InverseCamera. Do each photo one at a time.
• Complete project in single photo or multi-photo mode
• If using multiple photos, just complete the project as any normal PhotoModeler project.
• If single photo, map out planar sections and then use surface drawing or orthophoto. Note: the accuracy of surface draw or orthophoto depends very heavily on the solution of the overall project (i.e.. where the camera station ended up relative to the plane). If the orientation and project solution is not very accurate the surface draw or the orthophoto will be inaccurate also.
Note: there are some limits to using the single-photo axes constraints for solving camera parameters. First, format size aspect is not solved (whether it is checked or not in the Photo Properties). Second, you have to have strong three point perspective to solve for principal point. Third, you have to have at least a strong two point perspective to solve for focal length. If the perspective is weak or non-existent, the Audit Summary dialog will warn you before processing.