Journal of Autonomous Intelligence

Article: PC-VINS-Mono: A Robust Mono Visual-Inertial Odometry with Photometric Calibration

Published on
Updated on

Content:Feature detection and tracking rely on image gray value information, which is a very important process of visual inertial ranging (VIO). The tracking results significantly affect the accuracy of the estimation results and the robustness of VIO. In high-contrast lighting conditions, images taken by auto-exposure cameras often change with exposure time; gray values of the same features in the image change between frames, which makes the feature detection and tracking process very large The challenge; and the nonlinear camera response function and lens attenuation further exacerbate this problem. However, few VIO methods make full use of photometer camera calibration and discuss the effects of photometric calibration on VIO.

Yao Xiao, Xiaogang Ruan and Xiaoqing Zhu performed inertial measurements on the robust vision of the PC-VINS-Mono for photometric calibration, which was published in the journal Autonomous Intelligence.

In this measurement, the research team used the photometric response function, vignetting and exposure time to propose a robust visual inertial measurement of PC-VINS-Mono with photometric calibration. The proposed algorithm can be understood as an extension of the VENS-Mono with photometric calibration. With this extension, the proposed algorithm can be used in high contrast lighting conditions with automatic exposure cameras where the exposure time varies from frame to frame, which violates the brightness consistency assumption. The team evaluated the algorithm using the TUM VI dataset, which included different sequences under different lighting conditions. Comparative experiments show that the performance of PC-VINS-Mono is significantly improved by photometric calibration. For cameras with unknown response functions and lens attenuation coefficients, experimental results show that even with only exposure time calibration, the performance of our algorithm will increase in most cases. The experimental team skipping the CLAHE step showed that the performance of the algorithm was degraded, which confirmed that the CLAHE algorithm must be applied to improve the contrast of the image before feature detection, because photometric calibration may reduce the contrast of the image.

Click here to read more: