This master's thesis addresses the problem of loop closure detection in visual SLAM. Two approaches using binary features are presented: 1) A probabilistic approach called FAB-MAP is adjusted to use binary features for reliable loop closure detection. 2) A novel approach measures similarity between VLAD representations of image frames combined with pre-filtering to improve detection rate and speed. Both approaches are implemented and evaluated in the ORB-SLAM2 framework on the KITTI dataset, showing they achieve similar trajectory accuracy as ORB-SLAM2 while saving considerable time in loop detection compared to the original method.