車載カメラ運動時におけるFoveationの効果

—オプティカルフロー一様化モデルの構築—

抄録

In this paper, a virtual cylindrical screen (VCS) algorithm is proposed for detecting stationary objects inside a VCS when a sensor is moving in the forward direction along its optical axis. Simulations are carried out to investigate the advantages of foveation in the use of the VCS algorithm, and the advantages are discussed by comparing the pinhole camera (PHC) model with the sine curve (SC) model and the advanced wide angle foveated (AdWAF) model. A correction weight function is introduced for estimating the error in the optical flow that is calculated by using each image model. Moreover, the uniform optical flow (UOF) model for the VCS is developed for use in various applications, and an approximate image-height function is formulated for this model. The accuracy of the optical flow corresponding to a point on the VCS is investigated in simulations carried out using the correction weight function.