Automobile assisted driving system can not only reduce the driver′s mental stress during drivingdriving,but also greatly r educe the probability of traffic accidents.In order to improve the ability of dr iverless vehicles to perceive the external environment,this paper proposes a cas caded neural network framework to detect and classify road signs in virtual envi ronments.The framework combines a full convolutional neural network with an auxi liary structure and a modified classic LeNet-5network to perform traditional image processing using corrosion processing on th e edges of the extracted road sign areas and the generation of miscellaneous problems.The operator is optimized an d solved to realize the positioning and recognition of multiple types of road signs in various situations such as ra in and snow in virtual road images. Compared with the classical invariant moment feature,ORB global feature extract ion method,and YOLO,SSD artificial intelligence method,the proposed method has the advantages of high d etection accuracy and fast calculation speed.