Research On Illumination Robustness Of Monocular SLAM Visual Front-end

Complex and varied illumination makes computer vision researches difficult,especially for the simultaneous localization and mapping algorithm(SLAM).In recent years,SLAM systems can complete the task of accurate localization and mapping in general lighting environment.But in the low-light environment or environment with shadow,even the leading edge SLAM system often causes tracking problems due to the lack of illumination robustness of the visual front-end,and there are little research on this aspect.For the low-light environment,this paper first proposed an image feature extraction method based on double feature algorithm by improving the existing mature SLAM system to ensure that the system can extract enough feature information to complete frame matching in low-light environment.The performance of the back-end optimization thread is improved by bundle adjustment with two kinds of point features.And for some images acquired under the extreme dark environment,we designed an image frame preprocessing step based on brightness compensation to ensure the normal operation of the system.Experiments on open datasets show that the proposed algorithm improved the robustness of SLAM vision front-end in low-light environment to some extent,and retains its excellent performance in normal light environment.For the illumination robustness in environment with shadow,we also improved the visual front-end.We designed a tracking method that can effectively resist the interference of shadow.By detecting the shadow in HSI color space,and using the adaptive shadow removal algorithm based on sub-region matching,most of the shadow in the image frame can be eliminated,and enough feature information can be extracted from the visual front-end.For the residual shadow boundary,we proposed an outlier filtering algorithm to minimize the influence of shadow on the inter frame correlation.We also used two open datasets to evaluate this part of method,and the results are in line with our previous expectations.Compared with traditional method,the algorithm in this paper is more robust to the shadow in natural environment.