Adaptive visual SLAM by composite tracking

我们的目标是提出视觉同步定位和映射（VSLAM）系统，该系统对图像序列中场景和视图的大变化具有适应性和灵活性。我们深入研究了几种特征跟踪方法，发现SURF特征跟踪是我们的VSLAM系统完成三维重建的合适选择。在此基础上，提出了基于快速图像索引的关键帧选择算法，提高了摄像机跟踪算法的效率。在实验验证和性能评估方面，目前最先进的运动结构系统已经在各种场景和相机条件下获得了具有挑战性的真实图像序列。

We aimed at proposing visual simultaneous localization and mapping(VSLAM) system which is adaptive and flexible to large changes of scenes and views in image sequences. We deeply studied several feature tracking approaches and figured out that SURF feature tracking is an appropriate choice for accomplishing 3D reconstruction with our VSLAM system. Further, we extended the efficiency of camera tracking algorithm using the key-frame selection algorithm based on a fast image indexing method. Experimental validations and performance evaluations w. r. t. the state of the art structure-from-motion system have achieved on challenging real image sequences acquired under various scene and camera conditions.