One-Class Learning Based Algorithm for the Freeway Automatic Incident Detection

According to the characteristic of the freeway, choosing three the most sensitive traffic parameters of AID(Automatic Incident Detection) as the feature vectors, an one-class classification based AID algorithm is proposed. The method establishes the regional distribution model of learning samples, and constructs the decision function in the feature space. When the detected data fall into the inner of the decision region, it will be judged as no incident happened, or else incidents happened. This method does not require any transcendental statistical hypothesis about the distribution of samples, even if the distribution of samples is non-convex and unconnected, it can gain better decision function. The experiment results show that the new algorithm can obtain a higher IR, and limit FIR effectively, so it is a new and potential method for AID.