Tracking in a complex visual environment

Abstract

We examine the tracking of 3-dimensional targets moving in a complex (e.g. highly textured) visual environment, which makes the application of methods relying on static segmentation and feature correspondence very problematic, even under specific assumptions about the trajectory of the target. From the system kinematics and optical flow formalism we derive a general criterion, the Tracking Constraint, which specifies the optimal camera reorientation as the one that minimizes the optical flow in the center of the camera in the sense of an appropriate metric. A correspondence-free scheme is devised which employs dynamic segmentation and linear features of the image in order to capture global information about the scene and bypass numerical differentiation of the image intensity.KeywordsOptical FlowTarget MotionLinear FeatureTracking SchemeCamera RotationThese keywords were added by machine and not by the authors. This process is experimental and the keywords may be updated as the learning algorithm improves.