Online Photometric Calibration of Optical Flow Visual-Inertial SLAM System

Recently, optical flow method is widely used in visual simultaneous localization and mapping (VSLAM). Based on the two assumptions of gray invariant hypothesis and spatial consistency in optical flow method, but the first assumption for auto-exposure videos cannot be established, because the camera will automatically adjust the exposure parameters to make the overall image brighter and darker. The variation of illumination can greatly affect the accuracy and robustness of VSLAM. To solve the light-sensitivity problem of visual-inertial SLAM (VI_SLAM) based on sparse optical flow method, a real-time photometric calibration algorithm was proposed. Adaptive compensation KLT is used to replace the traditional KLT optical flow tracking, and the response function of the camera, irradiance attenuation factor and exposure time are jointly optimized to correct the image brightness value, enhance the stability of front-end visual tracking, and improve the accuracy of sliding window optimization at the back-end. We tested the optimized system (VINS_photon) on large datasets and compared it with state-of-the-art SLAM. Experimental results show that the performance of VINS_photon has been evidently improved after the optimization of photometric parameters.