Vican:Very Efficient Calibration Algorithm for Large Camera Networks
Vican is one of the fastest algorithms to compute the position and orientation of large sets of cameras.
This process is critical in retail applications where cameras provide real time information about the in-store activity : maintain a geometrically correct representation of the 3D environment and enable automatic processes that detect events and keep track of products and customers.
Vican was designed “from the ground up” having in mind the typical operational context of retail camera networks: a human operator travels aroudn the store with a calibration object while the hundreds of cameras capture images.
Typical calibration objects are “cubes” with tags on its faces. The first task is to precisely calibrate the object (know the relative position and orientation of each tag precisely).
As the object moves, parts of it (individual tags) are viewed simultaneously by pairs of cameras. Leveraging the overlap between the images, VICAN integrates millions of observations through an extremely fast optimization process, resulting in a precise estimation of the position and orientation of each camera relative to a global coordinate system.
Visit our github repository and read our extended paper for futher details.
