Training systems for many emerging threats need to be highly flexible and involve complex, realistic scenarios. Two recent studies analyzed small boat swarming attacks and found that no adequate training systems exist, particularly since live-firing at multiple targets is impractical. Augmented Reality (AR) - compositing real environments and simulated objects - can overcome this training gap as it allows replacing real ammunition and targets with virtual rounds and boats. Recent advancements in AR address the generation and display of shadows and lighting effects from the virtual objects onto the real scene. However, creating maritime AR environments bears additional difficulties because of the ocean's dynamics, its reflective nature, and visibility constraints such as fog and haze.

This paper presents methods for creating realistic reflections of computer-generated ships on live ocean video. After mirroring the ship, custom GPU shaders are applied to the reflection to smoothly blend it with the background ocean video. To test the quality of the composite images, a preliminary user study was conducted in which the participants had to determine the authenticity of images that were either real or automatically augmented with varying levels of fidelity. The participants started accepting the composite images as believable under certain environmental conditions, showing that our methods produce believable composites.

We built our reflection model into the Augmented Reality Virtual At-Sea Trainer. AR-VAST trains deck officers in marksmanship and procedures in maritime environments that were previously prohibitive to training, such as harbors and sea states that do not permit "robo-skis" and "killer tomatoes."