Efficient Camera Exposure Control for Visual Odometry via Deep Reinforcement Learning

Shuyang Zhang,Jinhao He,Yilong Zhu,Jin Wu,Jie Yuan

2024-08-30

Abstract:The stability of visual odometry (VO) systems is undermined by degraded image quality, especially in environments with significant illumination changes. This study employs a deep reinforcement learning (DRL) framework to train agents for exposure control, aiming to enhance imaging performance in challenging conditions. A lightweight image simulator is developed to facilitate the training process, enabling the diversification of image exposure and sequence trajectory. This setup enables completely offline training, eliminating the need for direct interaction with camera hardware and the real environments. Different levels of reward functions are crafted to enhance the VO systems, equipping the DRL agents with varying intelligence. Extensive experiments have shown that our exposure control agents achieve superior efficiency-with an average inference duration of 1.58 ms per frame on a CPU-and respond more quickly than traditional feedback control schemes. By choosing an appropriate reward function, agents acquire an intelligent understanding of motion trends and anticipate future illumination changes. This predictive capability allows VO systems to deliver more stable and precise odometry results. The codes and datasets are available at <a class="link-external link-https" href="https://github.com/ShuyangUni/drl_exposure_ctrl" rel="external noopener nofollow">this https URL</a>.

Robotics,Computer Vision and Pattern Recognition

What problem does this paper attempt to address?

The problem that this paper attempts to solve is the stability problem of the Visual Odometry (VO) system in environments with drastic changes in lighting conditions due to the degradation of image quality. Specifically:

1. **The impact of lighting changes on image quality**: In complex lighting environments, existing camera exposure control methods are difficult to quickly adapt to rapid changes in lighting, resulting in over - exposure or under - exposure of images, thereby losing key information and affecting the performance and safety of the visual odometry system.

2. **Limitations of existing exposure control methods**:
- **Gradient optimization methods**: These methods rely on feedback mechanisms and have a delay problem. They are unable to adjust exposure parameters in a timely manner and are prone to producing oversaturated images in dynamic lighting environments.
- **Function fitting methods**: Although they have a fast response speed, there are large differences between adjacent frames and they consume a large amount of computing resources.

To solve these problems, the paper proposes a camera exposure control system based on Deep Reinforcement Learning (DRL), aiming to optimize exposure settings through intelligent agents to improve image quality and the stability and accuracy of the visual odometry system.

### Main contributions

1. **DRL - based camera exposure control solution**: The exposure control problem is decomposed into two subtasks, achieving completely offline DRL training without direct interaction with the real environment.
2. **Light - weight image simulator**: Developed according to the imaging principle, it significantly improves data efficiency and simplifies the complexity of DRL training.
3. **Research on reward function design**: By designing reward functions with different levels of information, the trained agents have different degrees of intelligence and can perform well in challenging scenarios.
4. **Experimental verification**: The experimental results show that this method not only improves the performance of visual odometry tasks but also achieves a faster response speed and less time consumption.

Through these improvements, the DRL framework proposed in the paper can manage camera exposure more intelligently under extreme lighting conditions, thereby significantly improving the performance and reliability of the visual odometry system.