Design And Implementation Of Vehicle Detection Technology Based On Multi-feature Cascade Classifier

Nowadays traffic growing, congestion and blockage has become increasingly serious.With the development of technology, Intelligent Transportation Systems (ITS) is used toimprove the management level of road traffic,which become the future direction oftransportation management. Vision-based road detection road detection is always a hot topicin the field, so it develop rapid. The accuracy, validity, timeliness requirements of algorithmhave been increased. Road detection method mainly use statistics、single Harr-like features orsingle Adaboost cascade classifier,which need a long time to learn, row detection rate andnot always adapt to problem. Paper studies the cascade classifier based on feature in vehicledetection of image, research mainly as follows:1. Proposed integrate of features. Introducing Histograms of Oriented Gradient(HOG)features into Harr-like feature set. In order to reduce the computational complexityof feature compute, a method of HOG dimensionality reduction is used. integralimage and integralhistogram are used to calculate Harr-like feature and HOGfeature,which greatly reducing the amount of computation.2. As the single cascade classifier based on Harr-like feature has long running、rowdetection, paper proposed a cascade classifier based on the combination of Harr-likefeature and HOG feature.3. Proposed a cascade classifier combine with SVM on the basis of characteristicscombination. The classifier use the result feature of the front layers of cascadestructure to train SVM classifier. So the finally classifier has more robust、fastertraining rapid、higher detection and more adapter to the problem4. Done experiment of the classifier using testing sample under the VS2008andOpenCV (Open Source Computer Vision Library).Finally, analyzing then experimentresult. Experiments show that a cascade classifier based on multi-feature can better detect thevehicle in different contexts. It can have a high detection performance and meetcertainreal-time requirements, which is suit to the video surveillance and intelligenttransportation.