Floating-Bagging-Adaboost ensemble for object detection using local shape-based features

We propose a novel learning algorithm, called Bagging-Adaboost ensemble algorithm with floating search algorithm post optimization, for object detection that uses local shape-based feature. The feature use the chamfer distance as a shape comparison measure. It can be calculated very quickly using a look-up table. Random sampling boosting algorithm is used to form an object detector. Floating search post optimization procedure is used to remove base classifiers which cause higher error rates. The resulting classifier consists of fewer base classifiers yet achieves better generalization performance. To demonstrate our method we trained a system to detect pedestrians in complex natural scenes. Experimental results show that our system can extremely rapidly detect objects with high detection rate. The learning techniques can be extended to detect other objects.