Multiple object detection and localisation system using automatic feature selection

Abstract

The problem of object detection in images and videos has been treated by a large number of researches. Many design factors degrade the solution of that problem, among these factors are the manual modelling of the object, manual feature sets selection, handcrafting architecture, classifier's learning algorithm selection and learning algorithm parameter adjustment. Here, a generalised object detection and localisation system is presented. It has the ability to learn the object model automatically. The feature selection is automated by adopting the Adaboost algorithm as a feature selection and meta-learning algorithm. The proposed system combines the cascade-of-rejecters approach with different weak classifiers and feature sets. Using the proposed system eases the evaluation of feature types and classification algorithms on different datasets. To maintain the system generality, low-level object-independent features such as Histogram of Gradients (HoG), Haar-like and modified Haar-like are used.