Research On The Illumination Robust Visual SLAM Algorithm

Nowadays,vision-based localization and mapping algorithm has become the focused research direction in the field of Simultaneous Localization and Mapping(SLAM).The trajectory estimation and environment map are constructed through the movement of visual-sensors-equipped robot without prior information in SLAM.The visual SLAM algorithm has the advantages of low sensor cost,small cumulative drift error,and the ability to build three-dimensional map.However,it is necessary for researchers to discuss the positioning and mapping performance in dynamic illumination.In order to improve the robustness of the visual SLAM algorithm,obtain the better positioning performance and construct the reliable three-dimensional point cloud map under the illumination challenge,the thesis mainly studies the visual SLAM algorithm based on the following aspectsFirstly,the localization and mapping principle of sparse direct method odometry DSO(Direct Sparse Odometry)are introduced.The photometric residual formula derivation process and loop closure detection based on the Bag of words model are respectively given.In order to improve the distribution and robustness of the visual dictionary,the feature screening strategy for the back-end is proposed in the Chapter II.The simulation is designed to show the performance of the proposed visual feature algorithm.Secondly,a real-time exposure compensation algorithm based on the camera photometric imaging model is introduced in the chapter III.The photometric parameters mainly include the response function,exposure time,and vignette factors.Then,the online exposure compensation algorithm based on the above parameters is introduced.The residual energy equation based on the visual features optical flow tracking is established to iteratively solve the photometric parameters and compensate the pixels intensity.The simulation is designed to show that the image exposure can be effectively compensated and the photometric parameters are dynamically adjusted in reasonable range.Then,the real-time exposure algorithm and the sparse direct method visual odometry are combined to update the photometric parameters and camera pose in the Chapter IV.The results of multiple simulations and experiments show that the proposed method has better positioning and mapping performance.Finally,a loosely-coupled visual SLAM algorithm is proposed to further improve the performance of the algorithm under the dynamic illumination in Chapter V.The geometric positioning of the indirect formulation is coupled with the marginalized keyframes of the sparse direct SLAM.The pose and the three-dimensional mapping of keyframes are corrected based on the pose and pixel priors.The simulation shows that the algorithm has better positioning and mapping performance on the illumination challenge dataset.The actual experiment verifies the robustness of the proposed loosely-coupled SLAM algorithm in dynamic illumination.