Appearance based object pose estimation using regression models

This paper presents an appearance-based approach for object pose estimation using least square regression models. We try to find the subspace that maps the object image data onto their pose data directly, and use it for object pose estimation. In the approach, we first obtain a pair of training data set, i.e., object images and their pose parameters. The objectpsilas appearance model can be derived from ridge regression of training data. The object pose estimation from currently observed image is carried out using this model. We also introduce the kernel methods to cope with the non-linearity underlying training data set. Experiments for pose estimation are conducted on two objects. Performance of our appearance models is discussed through the comparison with linear and non-linear regression models.