Optical Flow for Autonomous Driving: Applications, Challenges and Improvements
This work solves domain-specific problems for autonomous driving systems, with incremental improvements in training strategies and a novel framework for low-light conditions.
The paper tackles optical flow estimation for autonomous driving by addressing challenges in fisheye cameras and low-light conditions, achieving strong generalization to real-world fisheye data and significant performance boosts in low light.
Optical flow estimation is a well-studied topic for automated driving applications. Many outstanding optical flow estimation methods have been proposed, but they become erroneous when tested in challenging scenarios that are commonly encountered. Despite the increasing use of fisheye cameras for near-field sensing in automated driving, there is very limited literature on optical flow estimation with strong lens distortion. Thus we propose and evaluate training strategies to improve a learning-based optical flow algorithm by leveraging the only existing fisheye dataset with optical flow ground truth. While trained with synthetic data, the model demonstrates strong capabilities to generalize to real world fisheye data. The other challenge neglected by existing state-of-the-art algorithms is low light. We propose a novel, generic semi-supervised framework that significantly boosts performances of existing methods in such conditions. To the best of our knowledge, this is the first approach that explicitly handles optical flow estimation in low light.