Abnormal Motion Areas Detection for Advanced Driver Assistance System

摘要

This paper proposes a novel abnormal motion detection method for advanced driver assistance system (AMAD-ADAS) containing techniques of lane markings detection, motion areas detection and abnormality quantification model. First, lane markings are detected based on Hough transform. Later, a novel motion areas detection method consisting of multiple layers operation, multiple motions operation and multiple areas operation (termed MMM method) is approved to detect candidate motion areas, which is more suitable for real environment than the traditional appearance-based method that can only detect specified object, such as vehicles and pedestrians. Finally, an abnormality quantification model is estimated to quantify the abnormality of each candidate area obtained by MMM method. The AMAD-ADAS shows its robustness through many experiments on the Reinhard [1] datasets.