Augmented reality applications require 3D model of environment to provide even more realistic experience. Unfortunately, however, most of researches on 3D modeling have been restricted to an offline process up to now, which conflicts with characteristics of AR such as realtime and online experience. In addition, it is barely possible not only to generate 3D model of whole world in advance but also trasfer the burden of 3D model generation to a user, which limits the usability of AR. Thus, it is required to draw the 3D model generation to an online stage from an offline stage. In this paper, we propose an online scene modeling method to generate 3D model of a scene, based on the keyframe-based SLAM which supports AR experience even in an unknown scene by generating a map of 3D points. The scene modeling process in this paper is a little computationally expensive in terms of real-time but it doesn't restrict real-time property of AR because it is executed on a background process. Therefore, a user will be provided with an interactive AR applications that support interactions between the real and virtual world even in an unknown environment.