A Method for Detection and Tracking of Moving Objects in an Industrial Environment using Stereo Vision

Abstract

The motion of three-dimensional objects in an industrial environment is monitored using a sequence of stereo-images taken from the same point of view. Mechanical vibrations of the camera system cause an apparent motion of objects in space, which can be compensated for by determining the relative rotation and translation of images captured at different time instances. Images are partitioned into square sections, and section correspondence is sought between consecutive frames in order to establish the image rotation and translation parameters. Tite segmented image disparity between consecutive frames determines the image area over which actual motion is observed and simplifies the tasks of image segmentation and obtaining the correspondence of object features over time and also in a stereo image pair. The obtained space coordinates of object feature points are combined in order to estimate the object motion parameters and predict its future course. The method is computationally fast and can be used to track fast moving objects in a semi-structured environment.