Object Detection in Real-World Hazy Scene

Accurate object detection in the real-world hazy scene is very important to some potential visual task, such as video surveillance, smart city, autonomous driving and so on. This paper focuses on two research problems, which are build a synthetic dataset of analyze effect prior knowledge joint learning model scene. Two frameworks proposed knowledge-guided (KODNet) dehazing (DONet). In KODNet, statistical will be used guide general network learn features during training, makes detector better adapt special scenario. DONet can effectively solve problem structural detail missing color distortion caused by image dehazing, thereby realizing improvement objects accuracy The experimental results RTTS show that KODNet effective they achieve mAP 70.5% 66.6%.