A Dynamic Feature Selection Method for Vision Based Vehicle Recognition

Current mainstream vehicle recognition algorithms mainly depend on the synthesis of both appearance based and knowledge based features to identify the candidate objects. Whereas, because of the unpredictable complex noises in real world environments, the existences, quantification and the explanation for certain features are often ambiguous which makes current algorithm hard to fulfill the dilemmatic high sensitivity/accuracy restriction, and an improvement for a certain feature(or data sets) often leads to a degeneration for others. This paper introduces a probability based feature selection method which enables the dynamic feature selection and multigrain feature evaluation. The experiment result (for rear vehicle recognition) shows the proposed method is an efficient way to improve both the sensitivity and the accuracy rates without the degeneration phenomenon.