A method is presented for tracking general curved objects through 3-space, given a sequence of grey-level images. The explicit recovery of 3D features is avoided and results demonstrate the method to be stable, accurate and robust. The object model has two parts -a tracking model and a grey-level model; the former specifies which features of the object are tracked, and the latter determines the appearance of these features. The method assumes initial position is known. Visible features are tracked using correlation between their rendered appearance and the next frame, to give a set of disparities. These disparities are used to invert the perspective transform and give the new position of the object.It has recently been demonstrated that to track known, rigid planar-faced objects through grey-level image sequences it is unnecessary to extract 3D features from the images [1,2,3,4,5,6]. Such work has shown that the approach of tracking 2D features and then inverting the perspective transform to recover 3D object position is not only highly robust to image noise and partial occlusion [3], but can also be implemented in real-time given current hardware [4,5,6]. Much of the above work concentrates upon tracking planarfaced objects, and uses only minimal modelling. In Bray's work -tracking a rotating plug through a sequence of "dirty" low-resolution images -robustness is achieved by using a small set of well-defined line features [2]; Stephens relies upon multiple viewpoints to achieve robustness [4]. Since Stephens and Harris both deal with real-time issues, their models consist only of a 3D-point set.This paper demonstrates that more sophisticated modelling reaps benefits in terms of accuracy of results and the stability of the tracking routine. More precisely, a grey-level modelling approach allows correlation between the predicted appearance of the model and the actual appearance, giving disparity vectors that will act to reduce accumulating error in position. It is suggested that such grey-level models can be computed automatically (e.g. [7,8]) and rendered in real time. The expense of this is partly offset by savings when inverting the perspective transform (since more consistent error vectors yield faster iterative solutions). The paper also demonstrates that the 2D approach provides a robust method of tracking rigid curved objects. All that is required for such tracking is that there exists a set of 3D features defining a model, that these features are precisely located, and that their visibility can be determined. However, further grey-level modelling is very useful for tracking these features through images. Finally, the methods are demonstrated on an image sequence that shows a real image rendered onto a rotating sphere. The sequence allows the estimated path in rotation/translation space to be compared with the true path. The tracking algorithm used is described in detail in [3, Chapter 4]. The method is conceptually simple: project the model at the predicted position in the next frame to get a set ...