Real time object detection in images based on an AdaBoost machine learning approach and a small training set.

摘要

Learning machine approaches for real time object detection are frequently used in literature. One of the most successful applications was introduced by Viola [39], who used AdaBoost for face detection in images. Our interest is to build fast and reliable object recognizers in images based on small training sets. This is important in cases where the training set needs to be built manually, as in the case that we study, the recognition of a type of car, specifically, the Honda Accord 2004 from rear views. We describe a novel variant of the AdaBoost learning algorithm, which builds a strong classifier by incremental addition of weak classifiers (WCs) that minimize the combined error of the already selected WCs. We describe a set of appropriate feature types for the considered recognition problem, including a redness measure and dominant edge orientations. We proposed to pre-eliminate features which do not satisfy a cumulative error threshold. Compared to existing literature, we achieve the design of a real time object detection machine with the least number of examples, the least number of weak classifiers, and with competitive detection and false positive rates.