Haze Visibility Enhancement for Promoting Traffic Situational Awareness in Vision-Enabled Intelligent Transportation
Yu Guo, Ryan Wen Liu, Yuxu Lu, Jiangtian Nie, Lingjuan Lyu, Zehui Xiong, Jiawen Kang, Han Yu, Dusit Niyato
Abstract
Visual sensors are one of the most essential sensing devices for achieving the traffic situational awareness of vehicles and monitoring equipment, as they can provide richer and more comprehensive information than other sensors. Therefore, many visual signal-based intelligent technologies have been proposed to perform a variety of traffic management tasks autonomously, including object detection, recognition, tracking, vehicle navigation, etc. Nonetheless, poor weather conditions, such as fog, haze, and mist, cause formidable challenges for visual technologies applied in intelligent transportation. To lessen the impacts of poor weather conditions, we propose a dual attention and dual frequency-guided dehazing network for enhancing visibility in real-time. In particular, the proposed model adopts an attention mechanism and a novel frequency information fusion strategy to extract global and local features and adequately recover sharp high-frequency structures and low-frequency details. Extensive experiments have revealed that our technique is superior to the state-of-the-art methods in terms of visibility augmentation and accuracy improvement of high-level visual tasks under hazy conditions.