An original vehicle detection algorithm based on object proposals

In this paper, we presented an improved vehicle detection algorithm based on object proposals. In the training part, by using Selective Search algorithm, we firstly segment the vehicle areas in the sample set as positive examples, other regions as negative examples. Then PHOG (Pyramid Histogram of Oriented Gradient) features of the positive samples and negative ones after separately being labeled 1 or 0 will be sent to the SVM (Support Vector Machine) to be trained for generating the initial classification model. In the test stage, test sample set is applied to evaluate the model to identify the mistakenly classified ones, which are called hard examples. They will be sent to the SVM to be implemented second-training, and the model will surely be updated. In the recognition part, the test images are firstly segmented by Selective Search algorithm to generate the candidate regions and then the PHOG features of each proposal will be extracted, which will be sent to the SVM model to be predicted as label 1 or 0 and the proposals whose features are predicted as label 1 will be reserved. The experimental results show that the vehicle recognition rate using the algorithm we proposed can achieve 97.52% and it performs favorably against existing state-of-the-art methods.