Freeway traffic incident detection using fuzzy CMAC neural networks

We present a new approach of incident detection based on a novel network architecture called the Fuzzy CMAC, and a feature extraction pre-processing algorithm using the nonlinear Karhunen-Loeve (K-L) transformation. We prove that the Fuzzy CMAC architecture is an excellent universal approximator that is able to learn an arbitrary traffic pattern discriminating function to any degree of accuracy with enough learning cycles. The learning rates are at least an order of magnitude faster than popular neural networks such as the multilayer perceptron. The nonlinear K-L transform proposed is able to aggregates the data collected directly from field detectors into a feature vector with much smaller dimensionality.