Introducing New AdaBoost Features for Real-Time Vehicle Detection
Bogdan Stanciulescu, Amaury Breheret, Fabien Moutarde · arXiv (Cornell University) · 2009
This paper shows how to improve the real-time object detection in complex robotics applications, by exploring new visual features as AdaBoost weak classifiers. These new features are symmetric Haar filters (enforcing global horizontal and vertical symmetry) and N-connexity control points. Experimental evaluation on a car database show that the latter appear to provide the best results for the vehicle-detection problem.