Abstract: A new visual perception computing model based on attention dynamics is proposed and applied to the auto-driving of a micro smart car. The model can select a visual attention region by combining the top driving task, bottom object saliency and prior knowledge. First, the task-related region is assigned by the top driving task； secondly, in the task-related region, the visual attention region is determined by a bottom object saliency detection method based on the modified Itti-Koch saliency map approach and an attention tracking and transferring mechanism based on prior knowledge. Finally, the recognition and perception process is executed in the visual attention region. Perception results are applied to the control decision, stored as prior knowledge and used to drive the task switching process by a feedback mechanism. The experiment results demonstrate that the new model can reduce the image computing time and improve the environmental perception ability and intelligent decision efficiency of the smart car.