Active tracking of foveated feature clusters using affine structure

Abstract

We describe a novel method of obtaining a fixation point on a moving object for a real-time gaze control system. The method makes use of a real-time implementation of a corner detector and tracker and reconstructs the image position of the desired fixation point from a cluster of corners detected on the object using the c&fine structure available from two or three views, The method is fast, reliable, viewpoint invariant, and insensitive to occlusion andlor individual corner dropout or reappearance. We compare two-and threedimensional forms of the algorithm, present results for the method in use with a high performance head/eye platform, and compare the results with two naive fixation methods.

Combining object and feature dynamics in probabilistic tracking

“…In the absence of high-level information, the best that can be done is to compute not only feature values but also uncertainty about the measurement. For ex- ample, dissimilarity computations [11] and Kalman filtering [10] have been used to estimate uncertainty of feature-point tracking. Kalman-filter based approaches such as MHT [2] and JPDAF [3] have been used to disambiguate feature tracks based on individual motion models.…”

Reactive Zoom Control while Tracking Using an Affine Camera

“…Tracking in our work is achieved using affine transfer [9,2], a method which takes advantage of the viewpoint invariance of single image features and the collective temporal coherence of a cloud of such features, without requiring features to exist through entire sequences. The method is fundamentally invariant to zoom [4,5], and thus independent of errors in zoom and whether the zoom control is reactive or purposive.…”

“…Overlaying the zoom-variant control process on top of a zoom-invariant tracking competence seems attractive from an architectural standpoint. The optimisation with respect to the Frobenius norm required to achieve transfer with more than the minimal point set was shown in [9] to be identical to Tomasi and Kanade's factorisation method [11], and so it is convenient to use the latter's standard formulation.…”

Affine-structure-based facial image encoding

“…We associate a constant image velocity Kalman filter [14,17] with each control point and each corner to be tracked. The state vector for each Kalman filter is given by The Kalman filter for each feature is initialised using the correspondence information obtained from the first three frames [14].…”