In the environment perception technology of autonomous vehicles, the dynamic sensing and tracking of vehicles are basic and key step. In order to improve the efficiency of vehicle perception in the traffic environment, this paper proposed a smart terminal car license plate perceptual method. The Mobilenet-SSD model is used to locate the position of the position of car first, then the position of the car license plate, finally, the OCR algorithm is used to recognize the character information on the car license plate. The experiment found that using EDSR preprocessing method to increase the resolution of image for poorly-captured car license plates can also improve the effective perception rate of the perceptual model. The car license plate in low-light scenes can significantly improve .At the same time, the confidence of perception of the car license plate in low-light scenes can be improved significantly. Under the condition of low confidence of 0.749, the information of the character of the car license plate can be perceived effectively. In addition, the transmission and feedback of video frames can effectively reduce the computing cost of vehicle sensors and improve the level of environmental perception and understanding of autonomous vehicles.

CICTP