Octree-Based Repetitive Pose Detection of Large-Scale Cyclic Environments

In order to achieve the pose detection in real-time by eliminating delay in large-scale cyclic environments, this paper presents an octree-based repetitive pose detection approach. Compared with the traditional detection method, the novelty of our algorithm lies in the high accuracy, in particular, consuming time does not rise with the expansion of the storage of poses. Based on the octree, the 3D space of position is segmented and converted into a binary format for storage or search. The orientations are stored at the leaf nodes of the octree, and the similarity of them is determined by the vectors matching method. The proposed algorithm is evaluated using real-world experiments and different simulated datasets including KITTI and Oxford, the evaluation results indicate the superior performance in terms of efficiency and accuracy using our algorithm in the large-scale cyclic environments.