Matching Images Features in a Wide Base Line with ICA Descriptors

In this paper we present a method to recognize images features with a wide base line between learning and recognition phases. The method is based in feature descriptors derived from independent component analysis (ICA). This technique is inspired by the problems of mobile robot mapping and localization using single camera. In the learning phase the descriptors are created to capture the variations in the appearance of each feature across a small base line tracking and stored in a database. The recognition phase proceeds to match descriptors created from the incoming video (with a wide base line respect to the learning phase) in the database. The implementation shows good computational performance