We propose a HMD (head mounted display) -based wearable AR (augmented reality) system that switches details of annotation information according to user's body motion such as walking and running, as a first step toward a more powerful and flexible context-aware AR system. Our body motion recognition algorithm uses a pair of accelerometers attached on user's thighs and recognizes five types of body motion -sitting, standing, walking, running and biking- as a user context by using a support vector machine. Through experimental studies with a prototype system, it was confirmed that the main concept of body motion-based information detail switching is widely accepted and that the recognition algorithm mostly functions as expected, though a few issues were identified that need to be improved such as a low running state recognition rate.