| 摘 要: 针对ORB-SLAM3(Oriented Fast and Rotated Brief Simultaneous Localization and Mapping)在外界光照条件变化时算法检测的特征点数量急剧减少且分布过于集中的问题,文章提出一种基于改进自适应阈值及四叉树的ORB_SLAM3算法。该方法首先通过计算特征点周围像素灰度值与平均灰度值的方差获得环境对比度;其次通过对比度计算光照自适应阈值;最后通过优化四叉树方法管理特征点,使图像中特征点的分布更加均匀。实验结果显示,正常光照强度条件下的特征点数量增加了9.5%、特征点均匀度提高了88.3%;低光照强度条件下的特征点数量增加了30.4%、特征点均匀度提高了34.0%;高光照强度条件下的特征点数量增加了72.9%、特征点均匀度提高了37.8%。以上说明该改进算法在视觉导航中具有较高的应用价值。 |
| 关键词: 特征点提取;光照自适应;优化四叉树;特征点数量;特征点均匀度 |
|
中图分类号: TP391.4
文献标识码: A
|
|
| Research on ORB_SLAM3 Algorithm Based on Improved Adaptive Threshold and Quadtree |
|
ZHANG Wei, HU Xuxiao
|
(College of Mechanical Engineering, Zhejiang Sci-Tech University, Hangzhou 310018, China)
616385569@qq.com; huxuxiao@zju.edu.cn
|
| Abstract: In view of the ORB-SLAM3 (Oriented Fast and Rotated Brief Simultaneous Localization and Mapping) algorithm's sharp decrease in the number of detected feature points and their excessive concentration under varying lighting conditions, this paper proposes an ORB-SLAM3 algorithm based on an improved adaptive threshold and quadtree. This method first calculates the environmental contrast by determining the variance of pixel gray values around the feature points in relation to the average gray value. Next, an adaptive threshold for lighting is computed based on this contrast, and finally, an optimized quadtree method is employed to manage the feature points, leading to a more uniform distribution of feature points within the image. Experimental results show that under normal lighting conditions, the number of feature points increases by 9.5% , and the uniformity of feature points improves by 88.3% ; under low lighting conditions, the number of feature points increases by 30. 4% , and the uniformity improves by 34.0% ; under high lighting conditions, the number of feature points increases by 72.9% , with a 37.8% improvement in uniformity. These results indicate that the improved algorithm has significant application value in visual navigation. |
| Keywords: feature point extraction; lighting adaptation; optimized quadtree; feature point count; feature point uniformity |
