In Augmented Reality (AR) systems, accurate and stable detection of printed AR markers have a significant impact on the quality of digital content superimposed in real space. Conventional AR marker detection methods mainly use images in the visible light range, which makes the detection performance susceptible to degradation due to changes in lighting conditions, shade, strong backlighting, complexity of the background and variations in print quality. To solve this problem, we propose a multispectral camera and image creation method that uses near-infrared and ultraviolet channels in addition to the visible light channel. The proposed method combines multiple images captured in different wavelength bands and takes advantage of the unique reflectance characteristics of each channel to reduce the influence of ambient light and background noise. By experiments, the effectiveness of the proposed system was verified under various imaging conditions, including shaded, direct sunlight, indoor light, near a window and partially shadowed environments. The experimental results show that the proposed system can stably detect AR markers in indoor and outdoor light environments and can also recognize them from a distance.