Domain Adaptive Multitask Model for Object Detection in Foggy Weather Conditions

Object detection in foggy weather conditions is a challenging problem, facing image degradation and feature space mismatch. A well-known method is to use image restoration methods to enhance degraded images before object detection. However, due to the limited availability of real-world datasets, these methods are based on synthetic datasets and have limited effectiveness in real-world scenarios. To address such issues, this paper proposes a domain adaptation-based multitask foggy weather object detection method, aiming to improve object detection performance in real-world foggy images. Firstly, we design an efficient Mini-AOD-Net as an image restoration network, which effectively enhances image clarity and preserves clean features for the detection network. Secondly, we introduce a domain adaptation module to address the object detection generalization problem in real-world scenarios. Experimental results demonstrate varying degrees of improvement in detection accuracy for both synthetic and real datasets using our proposed method. Furthermore, our approach can achieve a detection speed of 5.2 frames per second on an embedded system.