Enhancing Depth Image Estimation for Underwater Robots by Combining Image Processing and Machine Learning

Quang Truong Nguyen,Thanh Nguyen Canh,Xiem HoangVan

2024-11-08

Abstract:Depth information plays a crucial role in autonomous systems for environmental perception and robot state estimation. With the rapid development of deep neural network technology, depth estimation has been extensively studied and shown potential for practical applications. However, in particularly challenging environments such as low-light and noisy underwater conditions, direct application of machine learning models may not yield the desired results. Therefore, in this paper, we present an approach to enhance underwater image quality to improve depth estimation effectiveness. First, underwater images are processed through methods such as color compensation, brightness equalization, and enhancement of contrast and sharpness of objects in the image. Next, we perform depth estimation using the Udepth model on the enhanced images. Finally, the results are evaluated and presented to verify the effectiveness and accuracy of the enhanced depth image quality approach for underwater robots.

Robotics

What problem does this paper attempt to address?

This paper attempts to solve the problem that directly applying traditional machine - learning models for depth estimation has poor performance in low - light and noisy environments, such as underwater environments. Specifically, the paper focuses on how to improve the depth - estimation accuracy of underwater robots in complex environments.

### Main problems
1. **The particularity of the underwater environment**: Due to light absorption, scattering, and interference from living organisms and other objects in the underwater environment, the image quality is poor (such as color distortion, low contrast, etc.), which poses a challenge to traditional RGB - image - based depth - estimation methods.
2. **Limitations of existing methods**: Traditional methods such as sonar and laser scanners can be used for depth estimation, but they are greatly affected by noise and have low efficiency in the underwater environment. Although methods based on monocular cameras can provide dense depth information, they also face problems of insufficient data and complex processing in the underwater environment.

### Solutions
To address the above problems, the paper proposes a new method that combines image processing and machine learning, which mainly includes the following two contributions:

1. **Image - quality improvement**:
- A method based on color compensation, brightness balancing, and contrast and sharpness enhancement is proposed to improve the underwater image quality. By adjusting the color and brightness of the image, the image becomes clearer and has higher contrast, thus providing better input for subsequent depth estimation.

2. **Application of the depth - estimation model**:
- The Udepth model is used to estimate the depth of the improved image. Udepth is a monocular depth - estimation model based on deep learning, which can extract dense depth information from RGB images. This model combines the MobileNetV2 backbone network, m - Vision Transformer, and interpolation convolution blocks to achieve efficient and accurate depth estimation.

### Experimental results
Experimental verification shows that this method performs well on the USOD10K dataset. Compared with other methods (such as WaterNet, histogram equalization, and Unsharp Marking), it has significant improvements in multiple evaluation metrics (such as Abs Rel, Sq Rel, RMSE, and log10). This indicates that this method can effectively improve the accuracy and reliability of underwater depth estimation.

### Summary
The method of combining image processing and deep learning proposed in this paper successfully solves the challenges faced by depth estimation in the underwater environment and provides underwater robots with more accurate environmental - perception capabilities. Future research will further optimize the model to achieve real - time applications and improve the ability to adapt to different underwater environments.