This paper mainly focuses on an improved A* path planning algorithm for vision-guided automated guided vehicles(AGVs). A type of graphic coded mark method is firstly designed, which includes navigation, positioning as well as task information and ensures that the AGV can recognize the orderly arranged coded marks in the path to quickly locate and predict the next position, which lays the foundation for multi-AGV planning. Then, an improved A* algorithm considering the dynamic time cost is proposed to guarantee the path planning and collision avoidance strategy of multi-AGV in the path network, and effectively improve the utilization efficiency of path network resources. Finally, numerical simulation results show that the proposed improved algorithm can be effectively applied to the multi-AGV system and improve the entire work efficiency.