A LINE-MOD-based markerless tracking approachfor AR applications

Markerless tracking is still a very challenging problem in augmented reality applications, especially the real elements are textureless. In this paper, we proposed a model-based method to tackle the markerless tracking problem. Motivated by LINE-MOD algorithm, one of the state-of-the-art object detection methods, and multiview-based 3D model retrieval approach, we built a camera tracking system utilizing image retrieval. In the off-line training stage, 3D models were used to generate templates automatically. To estimate the camera pose accurately in the online matching stage, LINE-MOD was adapted into a scale-invariant descriptor using depth information obtained from Softkinetic, and an interpolation method combined with other mathematical calculations was used for camera pose refinement. The experimental result shows that the proposed method is fast and robust for markerless tracking in augmented reality environment; the tracking accuracy is much closer to that of ARToolKit markers.