An Modified Real-time Photometric Calibration Algorithm Based on ORB Features*

Abstract

In order to improve the accuracy of direct visual odometry and SLAM algorithm under illumination changes, we propose an modified real-time photometric calibration algorithm based on ORB features. Firstly, we use more robust ORB features as input to a optimization framework to obtain scene points correspondences. Then, we establish the reprojection pixels intensity residual equation based on photometric image formation and feature matching results of the camera. Finally, the incoming frames are corrected by the camera response function, exposure time and vignetting estimation. The proposed approach is tested with the EuRoC dataset and TUM dataset respectively, and experimental results indicate that the exposure compensation and time cost performances of the proposed method are promising.