基于改进关键帧选择的RGB-D SLAM算法

关键帧选择是提高视觉SLAM(simultaneous localization and mapping)算法精度及实时性的重要因素.关键帧常以图像的帧间相对运动距离为选择依据.该方法虽简单有效,但实时性、鲁棒性较差且容易产生大量冗余关键帧.针对上述问题,提出一种改进的关键帧选择算法.该算法整合了帧间相对运动距离、帧间特征点跟踪以及最小视觉变化来选择关键帧并删除冗余关键帧.基于该算法,结合具有较好方向和光照不变性的ORB(oriented FAST and rotated BRIEF)特征,实现了RGB-D SLAM算法.在RGB-D数据集上的实验表明,改进的关键帧选择算法能够更精准、及时地选择关键帧,并在减少RGB-D SLAM中冗余关键帧的同时提高算法的实时性、建图和定位精度.

RGB-D SLAM algorithm based on improved key-frame selection

The key-frame selection is an important factor to improve the accuracy and real-time of visual SLAM algorithm. The key-frame selection is often based on the relative motion distance of the images. This method is simple and effective, but with poor real-time and robustness, and is easy to generate a large number of redundant key-frames. Aiming at the above problems, an improved key-frame selection algorithm is proposed. The algorithm incorporates the distance of relative motion of the frames, the feature points tracking and the minimum visual change to select a key-frame and delete the redundant key-frames. Based on this algorithm, combined the ORB (oriented FAST and rotated BRIEF) features with better direction and illumination invariant, the RGB-D SLAM algorithm is implemented. Experiments on RGB-D datasets indicate that the improved key-frame selection algorithm can be more accurate, timely in selecting key-frames, and improve the real-time and the accuracy of mapping and localization while reduce the redundant key-frames.