Abstract

The estimation of velocity has been proposed as a preprocessing step for many high-level vision algorithms for digital images. A new Galerkin inverse finite element approach is described for estimating velocity efficiently which uses an adaptive triangular mesh in which the resolution only increases where motion is found to occur. The mesh facilitates a reduction in computational effort by enabling processing to focus on particular objects of interest in a scene, specifically those areas where motion is detected without expending effort on computing flow for regions in which there is no motion. A confidence is calculated for the detected motion and this motion is used for the adaptive mesh refinement process if its confidence passes a threshold. The method is applied to real images where a significant part of the image is static with results showing that the adaptive framework estimates motion efficiently without loss of accuracy.