OIPQ: A feature extraction algorithm based on ORB, Image Pyramid, and Quadtree Partition

The front-end feature extraction part of the ORB-SLAM2 algorithm extracts feature points through the use of a fixed resolution size grid for each layer of the pyramid image, causing the loss of some high-quality feature points and thus affecting the effectiveness of feature point matching. In the ORB-SLAM2, the feature selection method used is an approach based on the division of quadtrees, which suffers from a too-deep division, resulting in the poor real-time performance of the overall algorithm. To solve these problems, this paper come up with a feature extraction algorithm (OIPQ) based on ORB, Image Pyramid, and Quadtree Partition. It dynamically adjusts the grid size based on the number of feature points allocated in each layer of the image pyramid and also restricts the quadtree partitioning depth based on the number of feature points in each layer. These improvements aim to enhance the algorithm’s efficiency.