Wenhao Huang, Songyi Lu, Yifan Liu, Guoyin Zhang, and Quande Yuan
Visual odometer, deep learning, SuperPoint, feature point extraction
In view of the traditional monocular visual odometer in visual changes, light changes, poor robustness, low pose calculation accuracy, the feature matching module in ORB-SLAM replaced with feature matching based on SuperPoint network, and feature tracking, local map, key frame recognition, loop detection, pose estimation. Comparing the improved algorithm with the traditional ORB and SIFT on the public dataset KITTI, the absolute trajectory error was somewhat reduced, indicating that the method of integrating deep learning feature points is significantly better than the traditional visual SLAM in accuracy.
Important Links:
Go Back
