Moving object extraction with cooperative cameras

Abstract

SUMMARYWe present a novel approach to robust extraction of moving objects with cooperative static and active cameras. The active camera, for usual broadcasts, is fixed in location but free to rotate and zoom. The static camera is a completely fixed panoramic camera viewing a whole scene. Precalibration and on-the-ground calibration are carried out to determine the intrinsic and extrinsic camera parameters. The two cameras work in synchronism, and the motion parameters (e.g., pan, tilt, zoom, focus) of the active camera are measured for each image frame. Background subtraction is applied to the image sequence of the static camera to extract moving objects, and the obtained moving objects are then transformed to the image sequence of the active camera using both calibrated camera parameters and camera motion parameters. The approach is robust and fast and has a number of other desirable features.

Recognition and tracking of object positions in image sequences using multiple‐viewpoint television broadcast images

“…R ti | is minimized using V avg and α avg calculated in methods (2) and (3), respectively. Here, due to various errors in the coordinates obtained for the same player from local-view images and the wide-view image, we cannot expect to find exact matches.…”