Homologous Twin Snakes Integrated in a Bundle Block Adjustment

Snakes are active contours in the 2d image plane. Starting from an initial estimate of a curve's shape and position they move to salient image features in an iterative process driven by minimization of an energy function. The twin snakes concept is an extension to traditional snakes. Twin snakes are designed for detecting two parallel contours simultaneously and are useful for line extraction in high resolution imagery. For that purpose the snakes' energy function is extended by an additional term formulating the attraction force to a curve which is parallel to the twin partner. They are more robust than traditional snakes regarding inaccurate initialization, because the attraction force makes the snake approach to the desired line from far away. The presented method is applicable to both closed and open lines. After extraction of the 2d lines from several photos homologous lines are identified and integrated in a bundle block geometry. The 3d free-formed object curves are derived in an established bundle block adjustment system by intersecting the rays defined by snakes and camera models. As a result the object is described by geometrically reconstructed spatial object lines. An example of close range photogrammetric object reconstruction by lines with high degree of automation is shown.