Research On The Method Of Rain Detection And Removal In Video Image

Computer vision system is now widely used in various industries,including video surveillance,visual tracking and navigation,intelligent transportation,entertainment industry,and so on.Indoor circumstance of computer vision system has been universally used and studied,while some outdoor conditions,such as rain,snow,fog,are still a challenging problem in computer vision systems.In this paper,we study the impacts of bad weather(i.e.rain)on the video quality and how to detect and remove the rain drops in the image.Aiming at the problem of degradation of image quality caused by rain weather,this paper proposed a new method of detecting and removing rain drops,which not only can effectively remove the rain drops in video images,but also can availably eliminate the obstacles caused by camera movement,and be suitable for all kinds of rain scenarios.To the problem that the effect of rain removal in existing algorithms is unsatisfactory due to the detection accuracy be not enough,this paper presents a detection methods based on a variety of rain characteristics.This method can improve the detection accuracy,and lay a good foundation for the removal of raindrops.The algorithm firstly detect the possible rain streaks in the video frame,using the spatial features of rain,such as the intensity and structure of rain streak,and then mark the detected areas.Secondly,the rain streak candidates are selected based on a Gauss distribution model.Finally,in the processing of block matching,the detected regions are improved by using the advanced temporal characteristics to make the detection result more accurate.This method ensures the accuracy of the rain detection,by using a variety of characteristics of rain synthetically.It can also detect the rain streaks with all kinds of intensities,and it will not be affected by camera motion.On this basis,this paper proposes a more effective way to remove the rain.After the rain detection,a non-rain block matching algorithm is used for each of the labeled pixels in the current frame.The algorithm is used for the adjacent frames to find the similar blocks with the rain pixel regions.Once analogical blocks be found,they will be reconstructed by using a temporal non-local mean filter to remove the rain.Finally,we perform a specific post-processing algorithm to enhance the visibility and eliminate flicker.Unlike previous methods,this algorithm,by using the method of spatial NLM filter to remove the rain streaks,preserved the image details and reduced the fuzzy phenomena.So it presents better visual effect in the subjective visual quality.At the same time,the proposed algorithm can achieve the purpose of removing the rain with only five temporal neighboring frames,and can effectively suppress the noise.