Skip to Main Content
AdaBoost based learning algorithms have been traditionally used for the detection of objects with a fixed geometric structure, for example human faces. The learning algorithms can learn the geometric structure common to all instances of the face, and latter use the learned geometric structure for face detection. In contrast to the existing approaches, we use AdaBoost based learning algorithm for the detection of low level objects which do not have any fixed geometric structure, for example edge- corners. In contrast to human faces, edge-corners can occur in any possible orientation, and therefore possess no common geometric structure. Learning of such objects with a learning algorithm is significantly harder than the learning of structured objects like faces. However, we find that AdaBoost based learning algorithm can learn and then detect the edge-corners with very high accuracy. In our experiments, the true positive rate up to 100% is achieved for a false positive rate of only 10%.