Learning to Control Camera Exposure via Reinforcement Learning
IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2024
(*Equal Contribution)
[Paper]
DRL-AE
Features
The first Automatic Exposure (AE) control method using Deep Reinforcement Learning (DRL).
Novel training environment, state, action, and reward design for AE control.
Instant exposure adjustment within 3-5 frames.
Real-time processing on a CPU device (~1 ms).
Superior performance on feature extraction, object detection, and more vision applications.
Training framework overview
Abstract
Adjusting camera exposure in arbitrary lighting conditions is the first step to ensure the functionality of computer vision applications. Poorly adjusted camera exposure often leads to critical failure and performance degradation. Traditional camera exposure control methods require multiple convergence steps and time-consuming processes, making them unsuitable for dynamic lighting conditions. In this paper, we propose a new camera exposure control framework that rapidly controls camera exposure while performing real-time processing by exploiting deep reinforcement learning.
The proposed framework consists of four contributions:
1) a simplified training ground to simulate real-world's diverse and dynamic lighting changes,
2) flickering and image quantity-aware reward design, along with lightweight state-action design for real-time processing,
3) a static-to-dynamic lighting curriculum learning method to gradually improve the agent's exposure-adjusting capability, and
4) domain randomization techniques to exceed the limitation of the training ground and achieve seamless generalization in the wild.
As a result, our proposed method instantly reaches the desired exposure level within five steps with real-time processing (~1 ms). Also, the acquired images are well-exposed and show superiority in various computer vision tasks, such as feature extraction and object detection.
Video & demo result in real-world scenarios
Zero-shot generalization in real-world driving scenario (campus)
Zero-shot generalization in real-world driving scenario (urban road)
Exposure convergence comparison 1 (Ours vs Camera AE)
Exposure convergence comparison 2 (Ours vs Camera AE)
Feature extraction comparison (SIFT)
RoI-aware exposure control (DRL-AE)
Publication
"Learning to Control Camera Exposure via Reinforcement Learning" [PDF]
Kyunghyun Lee*, Ukcheol Shin*, and Byeong-Uk Lee (*Equal contribution)
IEEE / CVF Computer Vision and Pattern Recognition Conference (CVPR) 2024
Bibtext
TBA.