Camera Pose Recovery

The camera pose recovery module is responsible for determining the position and orientation (pose) of the camera for each input image. Recall that the input to PhotoSketch consists of an overlapping set of images of the scene to be modeled. By recovering the camera pose for each input image, it becomes possible to stand in SketchUp in the same position and hold the camera in the same orientation as the photographer did when those images were taken.

Camera pose recovery consists of a three-step process:

  • Feature Extraction
  • Feature Matching
  • Floor Alignment

Salient features are first extracted from each photo. They are then matched across successive pairs of input images to track their motion across the images. This motion permits us to solve for the camera poses. Simultaneously, a sparse 3D point cloud is derived from the salient features of each input image. The sparse point cloud and the set of camera poses (depicted as frustums) are shown here.


Floor alignment is needed to manually align cameras and the point cloud with respect to a plane that is parallel to the ground. We assume that the building facades are perpendicular to the ground. Therefore, it is important to find the orientation of the ground plane accurately. Otherwise, the extrusion of the footprints or rooftops will not pass

through the building facades. Floor alignment begins by having the user select three or more correspondence points among the first two input photos. The example here demonstrates the results of a user matching three points among two images.