GNSS/Vision/INS tight integration navigation and positioning method based on feature enhancement in low-light environments

Abstract To address the challenges of low image contrast and distortion in low-light environments, which make it difficult to extract visual features, leading to high mismatch rates, and subsequently causing visual navigation positioning drift and degradation of the performance of global navigation satellite system (GNSS)/Vision/ inertial navigation systems (INSs) integrated systems. This paper proposes a tight-coupled global navigation and positioning method based on image enhancement. First, a multi-head self-attention image enhancement module is designed, which integrates lighting information, depth-separable convolutions, and channel attention mechanisms. A custom loss function is used to improve the quality of low-light images, achieving a peak signal-to-noise ratio (PSNR) of 23.36 dB, significantly enhancing the robustness of subsequent feature extraction and matching; Furthermore, based on the enhanced visual feature information, the GNSS, visual, and IMU raw measurements are fused to construct the FE-(feature enhanced Gnss vision inertial odometry) tight integration navigation model. Navigation experiments using EOROC and self-collected datasets demonstrate that the proposed method achieves average positioning root mean square error (RMSE) of 0.188 m, 0.192 m, and 0.246 m along the X, Y , and Z axes, respectively, with positioning residual mean values reduced by 12.3% and 25.66% compared to GVINS and VINS-Fusion, respectively, angular errors are consistently within 1.76°, and the proposed method provides a technical reference for autonomous navigation of payloads in low-light scenarios.