Multimodal Semantics Extraction from User-Generated Videos

Algorithm 1

Layout classification based on camera locations.

(1) Compute the average location of each camera over the whole time during which that camera has been recording.

(2) Apply optimization for fitting the average camera locations to an ellipse, and find the optimal parameters that define the best-fit ellipse.

(3) Evaluate the Euclidean distance between each average camera location and the best-fit ellipse.  If the average distance is less than a predefined threshold, then we assign the nominal value “Elliptical” to the camera locations pattern of the considered event.