Research on ORB–SLAM2 Deep-Sea Visual Localization Technique Based on Image Enhancement

Abstract

To address the poor performance of underwater visual localization caused by image degradation, this paper proposes an improved ORB–SLAM2 underwater visual localization scheme based on image enhancement. First, an improved MSR–CLAHE algorithm is presented, which integrates the global consistency of multiscale Retinex (MSR) with the local contrast enhancement capability of contrast-limited adaptive histogram equalization (CLAHE). Exponential mapping and multiplicative fusion are introduced to balance multiscale enhancement, while guided filtering and HSV correction are employed to suppress halo artifacts and restore natural color balance. Experimental evaluations on the EUVP, SUIM, and Jiaolong deep-sea data sets demonstrate that the proposed algorithm achieves the highest average UCIQE and UIQM scores, significantly improving image clarity, contrast, and color fidelity. Subsequently, an improved version of the ORB–SLAM2 algorithm based on the proposed MSR–CLAHE algorithm is proposed. Experimental results of deep-sea visual localization show that the proposed method greatly increases the number of matched feature points and enhances both trajectory tracking accuracy and system robustness compared with the original approach. In summary, the improved ORB–SLAM2 underwater visual localization scheme based on image enhancement proposed in this paper can effectively track the motion trajectory of underwater robots, meeting the requirements for accurate and robust underwater autonomous localization.