Feature extraction algorithm based on improved ORB with adaptive threshold

The ORB algorithm extracts too many feature points in the case of insufficient illumination, the excessive clustering of feature points will lead to partial loss of information extracted by the system, which will increase the probability of incorrect matching of feature points. In order to solve this problem, this paper firstly proposes an enhanced graphical adaptive threshold of feature extraction ORB algorithm, using the enhanced algorithm to process the low light information collected from the front end, so that the blurred texture and corners are clearer. Secondly, in this paper, we propose an adaptive feature point extraction method with local threshold to improve the ability of ORB algorithm to extract feature points within the image area and solve the problem of image information loss. Furthermore, the homogenization algorithm is used to replace the traditional quad-tree algorithm to avoid the problem of excessive concentrated extraction of feature points, making the extraction of feature points of the image more uniform. Finally, the proposed algorithm is tested using a public dataset, the experimental results show that the improved algorithm outperforms similar excellent methods in terms of the number of feature points extracted and the correct matching rate, especially the matching rate of feature points is improved by 18.6% compared with that before the improvement.