Image enhancement method for visual localization and mapping in complex light scenarios

Shoujun Jia, Chun Liu, Hangbin Wu, Chen ChenIOP 2022 · DOI

Abstract

Visual localization and mapping have received considerable attention in the fields of computer vision, photogrammetry, and remote sensing. Image matching is key to visual localization and mapping. However, light is complex and uneven in real scenarios, creating difficulties in feature extraction and matching. Hence, feature mismatch and loss reduce the efficiency, accuracy, and robustness of visual localization and mapping. We developed a visual localization and mapping method for complex light scenarios based on image enhancement. Starting with initial images, the irradiance and reflectance components were separated based on logarithmic transformation. Our method strengthened high-frequency components and restrained low-frequency components with improved homomorphic filtering, restraining the light component and enhancing the important reflection component. The SIFT algorithm was used for feature detection and matching. The proposed method was tested on images with uneven light captured using a stereo vision camera in an indoor environment, focusing on visual localization and mapping. The experimental results emphasized that the method improved the rate of image localization and number of point clouds, as well as the reprojection error, which ranged from 0.85 to 0.82 on average. Thus, the proposed method is robust rather than probabilistic for improving visual localization and mapping under complex light conditions.

Introduction

Visual localization and mapping (SLAM) are fundamental to autonomous navigation and 3D reconstruction. These processes rely heavily on image matching—the ability to identify identical physical points across different camera frames. However, real-world environments often present "complex light scenarios," such as deep shadows or intense glares. Standard feature detectors like SIFT often struggle in these conditions, leading to feature loss in dark regions or "false features" caused by harsh lighting gradients. This results in poor localization success rates and high geometric errors.

Proposed Methodology

To address these challenges, this paper proposes an image enhancement pipeline that acts as a pre-processing stage for SLAM. The core innovation is an Improved Homomorphic Filtering technique designed to separate and process the physical components of image formation.

Flowchart of the proposed image enhancement and SLAM pipeline.

The process involves:

  1. Component Separation: A logarithmic transformation decomposes the image into irradiance (low-frequency lighting) and reflectance (high-frequency surface texture).
  2. Frequency Domain Processing: Using a Fast Fourier Transform (FFT), a filter function suppresses the low-frequency lighting components while amplifying the high-frequency textures essential for matching.
  3. Reconstruction: The signal is returned to the spatial domain via inverse FFT and exponential transformation.
  4. SLAM Integration: The enhanced images are processed using SIFT for feature detection and RANSAC for outlier removal, followed by bundle adjustment to optimize the 3D map.

Experimental Evaluation

The method was evaluated using a custom indoor dataset of 623 image pairs captured with a stereo vision system. The environment featured complex, uneven lighting on various surfaces. The proposed method was benchmarked against raw images and images processed with standard Histogram Equalization and Sharpen (HES).

Results and Discussion

The experimental results demonstrate that the proposed enhancement significantly improves SLAM performance in challenging lighting.

  • Localization Success: The Success Rate of Image Localization (SRIL) improved from 89.4% to 92.9%.
  • Precision: The mean Reprojection Error (RE) decreased from 0.85 to 0.82 pixels. In contrast, the HES method increased the error to 0.97, indicating that standard sharpening can introduce detrimental noise.
  • Matching Robustness: Under weak light, the proposed method maintained a Matching Precision (MP) above 88%, compared to ~83% for original images.
Comparison of the proposed method vs. HES across image pairs.

The results indicate that by treating lighting as low-frequency noise, the system becomes robust to shadows and highlights. Unlike standard methods that merely redistribute pixel intensities, this approach targets the underlying physics of the scene.

Conclusion

The proposed image enhancement method offers a robust "plug-in" solution for improving visual localization in complex lighting. While highly effective in static indoor environments, future research will focus on testing the method in dynamic scenes and optimizing the computational overhead of FFT processing for real-time robotic applications.

Summary updated

Abstract

Abstract Visual localization and mapping have received considerable attention in the fields of computer vision, photogrammetry, and remote sensing. Image matching is key to visual localization and mapping. However, light is complex and uneven in real scenarios, creating difficulties in feature extraction and matching. Hence, feature mismatch and loss reduce the efficiency, accuracy, and robustness of visual localization and mapping. We developed a visual localization and mapping method for complex light scenarios based on image enhancement. Starting with initial images, the irradiance and reflectance components were separated based on logarithmic transformation. Our method strengthened high-frequency components and restrained low-frequency components with improved homomorphic filtering, restraining the light component and enhancing the important reflection component. The SIFT algorithm was used for feature detection and matching. The proposed method was tested on images with uneven light captured using a stereo vision camera in an indoor environment, focusing on visual localization and mapping. The experimental results emphasized that the method improved the rate of image localization and number of point clouds, as well as the reprojection error, which ranged from 0.85 to 0.82 on average. Thus, the proposed method is robust rather than probabilistic for improving visual localization and mapping under complex light conditions.
BibTeX
@inproceedings{jia2022image,
  author = {Shoujun Jia and Chun Liu and Hangbin Wu and Chen Chen},
  title = {{Image enhancement method for visual localization and mapping in complex light scenarios}},
  booktitle = {IOP},
  year = {2022},
  doi = {10.1088/1755-1315/1004/1/012014},
}
Figure