基于在线光度标定的单目直接视觉SLAM

Monocular Direct Visual SLAM with Online Photometric Calibration

为了减少场景点的光照变化对直接视觉SLAM(simultaneous localization and mapping,SLAM)的影响,在稀疏直接法(direct sparse odometry with loop closure,LDSO)的基础上对其优化,结合光度标定来增强系统的整体性能。直接视觉SLAM,其图像配对的基本假设是建立在灰度一致性之上的,即同一个场景点在多个图像中是以恒定亮度值出现的。为了更好地满足这一假设,采用KLT(Kanade-Lucas-Tracker)来跟踪关键点,利用带光度参数的像素跟踪模型建立优化方程,优化序列的光度参数(曝光时间、晕影和响应函数),从而实现实时曝光补偿,增强了系统视觉前端的跟踪稳定性。最后,在公开数据集上进行实验。实验结果标明所提方法在部分序列上能有效地提高系统性能。

We propose a refined direct sparse odometry system with loop closure detection,which combines photometric calibration to enhance the overall performance of the system,to reduce the influence of the reflection illumination change of the object on the simultaneous localization and mapping(SLAM).Direct visual SLAM has a basic assumption of image matching,which is based on gray consistency;that is,the same scene appears in multiple images with constant brightness values.We use Kanade-Lucas Tracker(KLT)to track key points and establish an optimization equation by pixel tracking photometric parameters of the model(exposure time,vignetting,and response function)to satisfy this assumption precisely,thereby realizing a real-time exposure compensation and a robust front-end in tracking.Finally,we test the optimized system on the public datasets.Experimental results show that the system performance can be effectively improved on some datasets.