With the development of imaging and radar sensor technology, vehicle-mounted early warning systems can realize the recognition and tracking of pedestrians quickly and widely with higher accuracy. Compared with other warning methods, the Augmented Reality Head-up Display (AR-HUD) can provide real-time and accurate warning information for users and reduce the driving risk caused by the sight switch, so it has a strong application potential. Studies have revealed that the coding method of warning information has a significant impact on the effect of warning. However, the design standard of AR-HUD warning method has not been established yet, so there is of great significance to explore the coding method of AR-HUD for its application. With AR-HUD pedestrian warning as an example, the warning information was coded based on color, shape, flashing, and other design elements. The multi-modal quantitative analysis and evaluation of each coding method were carried out from the eye movement index, behavior index, and subjective index and the corresponding design suggestions were put forward. The color coding can significantly improve the identifiability of the code at the time of maintaining low failure rate and short response time, while non-color coding can enhance code identifiability by redundant coding. The shape coding will reduce the comprehensibility of coding but will reduce the failure rate and the flashing code can increase the failure rate and response time of warning. Subjectively, users believe that redundant coding is more warning than uni-dimensional coding. The research results can provide theoretical support for the interaction design of AR-HUD and the application of real vehicles.