Abstract: 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.