Efficient camera attribute adjustment for improved visual SLAM in extremely changing illumination environment

When the ambient illumination changes, the information loss of the image makes it difficult to realize visual simultaneous localization and mapping (SLAM). The existing camera attribute adjustment methods can alleviate the degradation of image quality, but they will fail when extreme illumination changes occur. To solve this problem, this paper proposes an efficient camera attribute adjustment method for improved visual SLAM, which can obtain the optimal exposure time and gain of the camera in the scene with extreme illumination changes. Firstly, a highly discriminative image evaluation metric is proposed based on the characteristics of visual SLAM, including entropy, enhanced gradient and brightness similarity, aiming at effectively distinguishing image quality when extreme illumination changes occur. On this basis, Bayesian optimization is used to adjust the combined exposure attribute. Rather than fully using synthetic images that may generate misguided prior data in the scene with extreme illumination changes, the real-synthetic image is used to ensure the correctness of the prior data, thus achieving robust and efficient Bayesian optimization. Finally, in order to reduce the adverse effects of image noise and motion blur on visual SLAM, an attribute allocation method is proposed to decompose the combined exposure attribute, so as to obtain the optimal exposure time and gain beneficial to visual SLAM. We collected experimental data in various indoor and outdoor environments with changing illumination, and verified the proposed method by using the classic visual SLAM system. The results show that compared with the baseline method, our method achieves more robust and efficient camera attribute adjustment, effectively improves the performance of visual SLAM in the environment with changing illumination, and can even achieve accurate pose estimation in the environment with extreme illumination changes.