An Evaluation of Different Extraction Methods and Support Vector Machine Kernels for Vehicle Type Classification

Traffic surveillance and monitoring is effective when the vehicles are accurately classified. It comes into effect when it is applied at toll booth centers, parking areas, security system, accident prevention etc. Several algorithms have been used for classification of vehicles till date. We use LBP, LDP and HOG methods in our paper to process the image which is taken at different angles at a fixed size (100*100 px) and extract vehicle’s feature information concerning their length, width and number of tyres, color, model to decide the type of vehicle. SVM classifier is used for the classification of the dataset. The VID dataset made by collecting various images is used for monitoring the processes. This paper compares the 3 feature extraction methods and concludes that HOG with SVM is the best among them which gives the highest accuracy of 95.3% to classify the type of vehicle.