Freeway Black Spot Identification Based On Statistical Pattern Recognition

Black Spot identification is one of the basic tasks in road safety management. The accuracy and effectiveness of identifying black spot is of a particular concern to the transportation management and design teams. With the recognition that various black spot identification techniques have been developed, this thesis attempts to pinpoint the black spots on freeways according to statistical pattern recognition method.The thesis starts with the introduction of static and stochastic factors which could potentially impact identifying black spots, and how these two factors could influence the accuracy of identification results. In the meanwhile, the thesis discusses the principles of black spot identification model based on pattern recognition and its associated pros and cons.With all the necessary experimental data collected, the thesis developes three statistical pattern recognition models based on Least Square Regression, Fisher's Linear Discriminant, .and BP Neural Network using Matlab. Justified with the analytical results, the research eliminates linear methods having higher Mean Square Error and utilizes BP Neural Network. In attempting to choose the most appropriate BP Neural Network, the research uses the basic algorithm and eleven modified neural network algorithms to train the collected data. After speed of statistical convergence and MSE (Mean Square Error) of these algorithms are compared, Bayes Generalization algorithm is assessed to be the most appropriate.The last section deploys the established models on the real freeway segments to identify the black spots and consequently an evaluation analysis is conducted based on the field implementation. Furthermore, the thesis states that the proposed black spot identification technique shall exclude the effects of stochastic variables, otherwise the technique can't demonstrate its effectiveness.