Vehicle Detection Based On HOG And Haar-like Features

In the world,the sharply increasing number of cars has caused various traffic problems, so people need Intelligent Transport Systems(short for ITS) to improve traffic efficiency. Intelligent Transport Systems is a service system for traffic management and transport which is based on Internet Platform. Vehicle detection is one important branch of Intelligent Transport Systems.The main contents are as follows:The paper introduces the traditonal methods of vehicle detection, including background subtraction, frame-difference and optical flow methods. In order to improve the accuracy of vehicle detection,the two-step strategy for vehicle detection:is adopted:(1) Hypothesis Generation;(2) Hypothesis Verification.In the Hypothesis Generation,firstly,vehicle images are preprocessse, using the median method to remove white noise, and then Otsu method is used to extract vehicle shadow feature, using Canny edge detector is used to extract the vehicle symmetrical feature.Finally,we use the two features to generate hypothesis vehicle region.In the Hypothesis Verification, the paper introduces the fusion of two types of features:the rectangular filters (Haar-like features) and the histograms of oriented gradient (HOG). AdaBoost cascade detector with fusion features are used for Hypothesis Verification. in the early layers of the cascade.classifiers with HOG can eliminate easy non-vehicle examples,while in the later layers, the classifiers with Haar-like generate a fine decision boundary removing none-vehicle near the vehicle model.The paper implemented the training and testing part using MATLAB software.The experiment result in the paper shows that our vehicle method is of good detection accuracy and computational efficiency.It could largely tackle the varied vehicle and complex background situation.