To solve the safety problem in the autonomous driving system, metamorphic testing was used to determine whether autonomous driving systems predicted inconsistent driving behavior, but the metamorphic relations involved in the previous method cannot reflect the complex driving scene at night in real life. The night driving test is necessary because of the limited vision and light effects at night, which lead to many traffic accidents. In order to solve these above problems, this paper proposes FakeNight for night driving scenes, which is a DNN-based autonomous driving system consistency test framework. Experiments are conducted on multiple autonomous driving systems by using BDD100K datasets and Udacity datasets to verify the validity of FakeNight. In the experiment, the authenticity of the synthesized pictures and fault detection ability are measured. The experimental results show that the framework can effectively synthesize the night driving scene and discover potential defects in the autonomous driving system.

metamorphic testing; autonomous driving system; metamorphic relation; night driving scene