Vehicle traffic density estimation using Bayes, rule, tree family data mining classifiers applied on background subtracted traffic images

摘要

This paper proposes an innovative approach to traffic density estimation. It defines a method that focuses on reducing computational time and complexity by extracting row, column and diagonal mean feature vectors from the image. Then these feature vectors are used to train classifiers and the images are classified as low, moderate or high traffic situations. The system works in 2 phases: Training phase and classification phase. In the training phase, the image is subtracted to obtain the vehicles. The features of the subtracted image are extracted and a dataset is created. This dataset is used to train the classifier. In the second phase, the trained classifier is used to classify the real-time traffic data. Finally, seven data mining classifiers are used along with total fifteen combinations of feature vectors to test the accuracy of the eighty-four variations of the proposed technique. The Bayes family is proved to be better for traffic classification. The column mean features have been proven better. Overall Naïve Bayes classifier with column mean feature vector has given the better accuracy among experimented data mining classifiers.