Research On Video Rain Removal Algorithm Based On Deep Learning

With The topic of this article stems from the school-enterprise cooperation project "Intelligent Cockpit",and the proposed rain removal algorithm is mainly applied to this system.The intelligent cockpit is mainly applied in the field of unmanned aerial vehicle driving,which is a new type of flight control system that is intelligent,digital,and automated.Rain removal technology has the advantages of automation,high efficiency,and accuracy,which can improve flight safety and efficiency,providing more complete technical support for the implementation and application of the intelligent cockpit.In the field of computer vision,removing rain from videos is an important problem.This requires the computer system to be able to restore the background of rainy day images.In traditional video de-raining methods,rain streaks can be divided into two dimensions: spatial and temporal.However,most algorithms only focus on the spatial dimension,primarily concentrating on removing rain streaks.These algorithms are trained only on synthetic data,ignoring more complex degradation factors such as rain accumulation,occlusion,and prior knowledge in real data.Therefore,this paper proposes a more comprehensive rain model with multiple degradation factors.A two-stage video de-raining method is constructed that combines synthetic and real videos,physical prior restoration guidance,and adversarial learning.This method accurately restores the background information without rain streaks.It adopts the ideas of physical priors and adversarial learning,encoding a large amount of physical restoration knowledge into the model as a loss function.Based on multi-frame samples,the model learns a very powerful and precise video de-raining algorithm.Specifically,the network is divided into two stages: restoration guided by the physical model,and further restoration through adversarial learning.In the first stage,the reverse restoration process is performed under the guidance of the proposed rain model.An initial estimated background frame is obtained based on the input rain frame.In the second stage,adversarial learning is used to refine the results,including the overall color and lighting distribution of the restored frame.Experimental results show that the proposed method can achieve high-speed,highavailability,and high-precision video de-raining.Compared to other existing methods,the proposed method has higher de-raining rate and lower noise rate,effectively improving the quality of rainy day videos.