Detecting expressway traffic incident by traffic flow and robust statistics

In order to detect expressway traffic incidents veritably, the probabilistic identifying results are suggested to demonstrate the traffic jam/congestion detection by traffic flow. First, the noise series in a traffic flow is decomposed by wavelet denoising and by second kind Fourier analysis denoising respectively. Second, hypothesis testing by chi-squared distribution is employed to sort out the robust variance (scale) estimators. Last, assuming that the detected flow point with possible congestion follows a normal distribution with the mean of the decomposed signal and the variance estimated from the decomposed noise series previously, the confidence levels of congestions are generated in a probabilistic form. Numerical experiments show that the probabilistic identifying results by this three steps method are reliable and reasonable.