ORB-SLAM describes sub-matching feature points by ORB (oriented FAST and rotated BRIEF). Its illumination intensity robustness is insufficient and it is difficult to apply in poor illumination conditions. For this reason, taking advantage of the strong robustness of hue illumination intensity in HSV space, a method of adding hue to ORB feature matching in front-end matching based on Gauss mixture model is proposed to solve the problem of insufficient robustness of illumination intensity in feature matching. Based on Bayesian filtering model, a visual dictionary is constructed according to the current scene to complete loop detection and improve the accuracy of SLAM algorithm after position and attitude optimization by Bundle Adjustment. The experimental results show that, compared with ORB-SLAM algorithm, the accuracy and robustness of illumination intensity of the proposed algorithm are significantly improved under the condition of guaranteeing the real-time performance unchanged.