Vehicle Detection Algorithm Based on Modified Gradient Oriented Histogram Feature

In this study, algorithm development was conducted in two steps. Step 1 focused on training the algorithm using positive and negative samples. To increase execution speed, principal direction was adopted as the first feature to be identified. Subsequently, vehicle regions were converted into modified histogram of oriented gradients format and entered as inputs to a support vector machine (SVM) to identify the second feature. In the vehicle detection process, the first feature was adopted to eliminate nonvehicle regions first, and SVM training results were used to identify actual vehicle regions. Experimental results indicated that the proposed algorithm can effectively detect vehicles with an accuracy rate of up to 98%; moreover, the proposed method was approximately 40% faster than an SVM-based detection algorithm using the features of histograms of oriented gradients.