Abstract:

In view of the problems of the unclear definition of scenes and their content， the extraction of basic scene elements is limited to subjective analysis and the selection and design of scene element of different test subjects are inexplainable etc.， an extraction method of the key scene elements for vehicle autonomous driving simulation is proposed. The method analyzes the effects of scene elements on the modules of perception， decision?making and motion?control from the perspective of autonomous driving system， and establishes a mapping equation of element?structure?function plane according to the effects of different sub?modules of autonomous driving. Then， an extraction model of scene elements is set up based on the plane node discriminant matrix， to quantify the importance of scene elements with discrimination and screening performed. Finally， through analyzing the element composition of the pedestrian collision avoidance scenes in the 4th World Intelligent Driving Simulation Challenge， the feasibility and effectiveness of the method proposed are verified.

Key words: autonomous driving system, simulation scenes, scene element extraction