基于优化特征点提取的视觉SLAM算法研究

Study on Visual SLAM Algorithm Based on Optimized Feature Point Extraction

近年来,伴随着人工智能的发展,SLAM算法的运用也逐渐出现在人们视野中。然而,现有的视觉SLAM算法大多只包含基本的处理框架,在细节部分的处理仍比较粗糙,尤其是特征点的提取部分。鉴于此,以开源代码ORB-SLAM2为基础,对基于四叉树的ORB算法(Quadtree-based ORB algorithm,QtreeORB)进行优化,进而提出一种改进的SLAM算法。基于QtreeORB算法的基础上,提出一种自适应阈值的特征点提取方法。限制四叉树网格划分深度的同时,对特征点数量不同的节点进行不同程度的迭代提取,再根据是否达到Harris响应值的阈值来筛选合适的特征点,从而避免了特征点的过度均匀化。通过在TUM数据集上的实践结果表明,该算法所提取的图像特征点质量普遍较高,成功地避免了特征点的过度均匀化。

In recent years,with the development of artificial intelligence,the application of SLAM algorithm has gradually appeared in people’s vision.However,most of the existing visual slam algorithms only contain the basic processing framework,and the processing of the detail part is still relatively rough,especially the feature point extraction part.In view of this,based on the open source code ORB-SLAM2,this paper studies the quadtree based orb algorithm(QtreeORB),and then an improved SLAM algorithm is proposed.Based on QtreeORB algorithm,this paper proposes an adaptive threshold feature point extraction method.At the same time,the nodes with different number of feature points are extracted iteratively at the same time,and the appropriate feature points are selected according to whether the threshold value of Harris response value is reached,so as to avoid excessive homogenization of feature points.The experimental results on the tum dataset show that the quality of the feature points extracted by the algorithm is generally high,and the over homogenization of the feature points is avoided successfully.