基于视觉点线特征自适应融合的VINS定位方法

VINS LOCALIZATION METHOD BASED ON ADAPTIVE FUSION OF VISUAL POINT AND LINE FEATURES

由于Vins-mono算法在弱纹理环境中的定位精度不高,直接引入线特征的PL-Vinsmono算法的实时性能较差,故提出一种基于视觉点线特征自适应融合的VINS定位方法。首先在前端评估点特征质量,然后根据点特征质量判断是否引入线特征,最后在后端构建不同的损失函数进行迭代,提高了算法在弱纹理环境中的定位精度,也提高了在丰富纹理环境中的实时性。实验结果表明:在弱纹理环境下,该算法比Vins-mono算法的平均定位精度提高了16.7%;在丰富纹理环境下,该算法比PL-vinsmono算法的平均耗时减少了41.7%。

Due to the low positioning accuracy of Vinsmono algorithm in weak texture environment,and the poor real-time performance of PL-vinsmono algorithm which directly introduces line features,this paper proposes a VINS positioning method based on adaptive fusion of visual point and line features.The point feature quality was evaluated at the front end,and the line feature was determined according to the point feature quality.Different loss functions were constructed on the back end for iteration,which improved the localization accuracy of the algorithm in weak texture environment and improved the real-time performance in rich texture environment.Experimental results show that the average positioning accuracy of the algorithm is 16.7%higher than that of Vinsmono algorithm in the weak texture environment,and in the rich texture environment,the average time consumption of PL-vinsmono algorithm is reduced by 41.7%.