Good Features to Track with Appearance Models

Parameterized Appearance Models (PAMs) (e.g. Eigentracking [2], active shape models [6], active appearance models [5, 12], morphable models [3, 11]) are commonly used to model the appearance and shape variation of objects in images. While PAMs have numerous advantages relative to alternative approaches, they have at least two drawbacks. First, they are especially prone to local minima in fitting. This problem becomes increasingly problematic as the number of parameters to estimate grows. Second, often few, if any, of the local minima correspond to the expected location of the landmarks. This paper proposes two strategies to improve fitting in appearance models. First, we define a new criteria to select optimal regions to track based on the statistical properties of the error surface. Second, we learn a subspace that maps the local error to an ideal error surface to fit appearance models. The effectiveness and robustness of the proposed algorithm is demonstrated in real data.