Over recent years, the interest in the use of higher-order models for motion estimation and tracking has been growing. One important reason for this is that such models have the potential of utilising larger analysis windows, thus achieving a greater noise resilience without sacrificing model accuracy. However, the price of this increased flexibility is often an increase in representational and computational complexity. This paper summarises work on a novel approach to motion analysis which seeks to provide the flexibility of a higher order model without incurring a significant increase in complexity. It is based on the use of local affine models incorporated within a multiresolution framework and local frequency domain methods as the computational tool for estimating the model parameters. The work has led to the design of robust and efficient motion estimation and region tracking algorithms. The motion estimation algorithm is based on a generalised form of correlation implemented via the frequency domain. Affine motion parameters are estimated by aligning region spectra prior to correlating using a centriod pair technique. The use of frequency domain alignment and correlation is both robust and computationally efficient and the algorithm is capable of measuring the affine distortion between local image regions even in the presence of severe transformation and noise corruption. The tracking algorithm employs a concise `mid-level vision' region representation, the G-blob representation, based on a hierarchy of 2-D Gaussians undergoing affine motion. This representation has the property that it is closed under affine transformation, thus forming an ideal `infrastructure' for tracking regions undergoing affine motion. The G-blobs are tracked using a combination of the affine motion estimator and a standard predict-update Kalman filter. The advantage and innovation of the approach is that motion is the scene can be described using very few elements, yielding a concise representational structure. After a brief overview of the background to the work, the paper will outline the affine motion and G-blob models employed and describe the estimation and tracking algorithms. Results obtained on both artificial and real scenes will be presented.