Research And Implementation Of Image Night Removal Techniques For Monitoring Scenes

The image captured at night is affected by external uncontrollable factors,so that the quality of it is often significantly reduced.Therefore,the goal of image night removal is to generate a daytime image with better quality and richer details under the same scene with the input of night image.Image night removal for surveillance scenes is a very worthwhile research topic,since it can enable more night images to be applied to computer vision tasks that rely on high-quality input,such as image segmentation,object tracking,pedestrian detection and so on.And at the same time,it can also meet the daily needs of people to convert poor-quality night images into better-quality and richer daytime images.In order to solve the difficulty of converting night images to day images due to low visibility and complex lighting conditions of night images,this paper proposes a dual-branch generative adversarial network DeNight-Net for image night removal which embeds an attention model.Specifically,the attention branch is used to identify areas with large differences in day and night images,and the corresponding attention feature maps are generated to act on the generation branch,so that the generation branch can achieve a more effective image night removal.Secondly,this paper constructs a large-scale paired image de-night dataset Denight-138 which can be used for image de-night algorithm research and quantitative evaluation.In addition,this paper proposes two strategies,namely unpaired training strategy and the strategy of structuring paired dataset,to ensure that the DeNight-Net still maintains good robustness of night-to-day translation in the absence of paired datasets in actual scenarios.Finally,based on the proposed algorithm,a WeChat mini program for image night removal is implemented to meet the daily needs.In order to evaluate the effectiveness of the proposed dual-branch image night removal algorithm based on generative adversarial network,and the stability and real-time performance of the image de-night WeChat mini program,a large number of experiments were performed.The results show that the algorithm proposed in this paper has better performance than the latest image de-night methods,and the system based on this algorithm has high efficiency and reliability.