Vision-Based Detection and Labelling of Multiple Vehicle Parts

This paper presents a method for the visual detection of parts of interest on the outer surface of vehicles. The proposed method combines computer vision techniques and machine learning algorithms to process images of lateral views of automobiles. The aim of this approach is to determine the location of a set of car parts in ordinary scenes. The approach can be used in the intelligent transportation industry to construct advanced monitoring and security applications. The key contributions of this work are the introduction of a methodology to locate multiple patterns in cluttered scenes of vehicles which makes use of a probabilistic technique to reduce false detection, and the proposal of a method for inferring the location of regions of interest using a priori knowledge. The results demonstrate excellent performance in the task of detecting up to fourteen different car parts over a vehicle.