A feature based method for real time vehicle detection and classification from on-road videos

Vision Based vehicle detection and classification has become an active area of research for intelligent transportation system. But this task is very difficult and challenging due to the dynamic condition of roads. Many solutions have been given by the researchers to overcome these challenges. Some of them are giving good performance but computationally highly expensive and fail in some circumstances. In the proposed method, a feature based cost effective detection and classification method is proposed that is suitable for real time applications, provide satisfactory accuracy and computationally cheap. The proposed method uses haar-like features of the images and AdaBoost classifier for detection which provides a very fast detection rate with high accuracy. To reduce inconsiderable false positive rate generated by this method, we propose to use two virtual detection lines (VDL) that reduces the false positive rate. In order to predict the class of a vehicle, a feature based method is proposed. HOG, SIFT, SURF all are well represented feature for image classification. The existing feature based vehicle classification methods lack accuracy because of using those features inefficiently. In order to reduce those lacking, we propose to use bag of visual words (BOVW) model for classification. BOVW model also needs a lower computation time and resources. As the proposed method aims to be implemented in real time, we propose to use SURF feature for BOVW which is faster to compute and well described for recognizing an object. The BOVW then used for identifying the vehicle's class by multi class SVM classifier. Error correcting output code (ECOC) framework is used to achieve multi class prediction with SVM. Extensive experiments have been carried out on different traffic data of varying environments to evaluate the detection and classification performance of the proposed method. Experiment results demonstrate that the proposed method achieves a significant improvement in classification of heterogeneous vehicles in terms of accuracy with a considerable execution time as compared to other methods.