Ding Wang, Jing Wang, and Huan Wang
VSLAM, matching, feature extraction, adaptive threshold, texture-less, structure-less
Visual simultaneous localization and mapping (VSLAM) is a key algorithm in the navigation of mobile robots to be studied by many researchers. In VSLAM, it is the first and fundamental key technology to determine the common features of the different views of the same object from the acquired environment images. It determines the adaptability of the VSLAM algorithm to difficult environments with poor textures and structures. Therefore, the challenge of the VSLAM algorithm imposed by the poor texture and structure of the indoor environment is indeed the requirement of the good matching technology. To this end, that two-stage frame matching in VSLAM based on feature extraction with adaptive threshold for indoor texture-less and structure-less is proposed in the paper. The ORB algorithm with fast extraction speed and strong real-time performance is improved, adding adaptability to the FAST corner detection algorithm while increasing the scale invariance, so that the algorithm can automatically adjust the threshold to get enough feature points. After comparing the common feature matching methods, the violent matching was selected as the coarse matching, and the improved mismatches elimination method based on the motion smoothing model is proposed. The experiments show the improved feature extraction and matching algorithm Ours can extract enough key points in texture and structure-free environment, refine the local points to achieve better feature matching between current frame and reference frame, and provide a basis for subsequent pose estimation.
Important Links:
Go Back
