Research On Multiple Features Based Vehicle Detection

Vehicle detection is a very important part of vehicle locating system in theparking lots, accurate vehicle detection can correctly label the current state ofparking spaces, and significantly reduce the load on the license plate recognitionsystem. Currently, the vehicle detection methods are mostly used in intelligenttransportation systems, the number of vehicle detection methods used in parkingslot is still relatively small. Moreover, the appearance-based methods are susceptibleto the change of light and view, dim light and multiple views increase the difficulty ofthe problem. Common methods are easily affected by illumination.This paper presents a car detection method for parking lots by detecting carfaces. First, our method removes the video frames with dramatic changes usingframe subtraction and reduce the influence of illumination by Gamma correction.Second, we select EOH and Haar features to detect car faces by AdaBoost. However,Due to the diversity of the appearance of vehicles, it is difficult to achieve satisfactoryresults by using car face detection method alone. Our vehicle detection methodfollows two basic steps: hypothesis generation and hypothesis verification, thelicense plate detection method based on HOG feature detects the candidate regionsthat the car face detection method generates, so the method can further improvethe recall rate while significantly increase the precision rate. In addition, by the priorknowledge of the fixed size of the license plate, our method can estimate the generalsize of car face to reduce the searching range of classifier and increase the precisionrate. Experiments show that the proposed method can achieve high recall rate andprecision rate, and the method can overcome some of the occlusion.