Research On Rain Removal In Video Image

Computer vision systems are becoming increasingly widely used in the field ofintelligent transportation, industrial production and target recognition. However,bad weather will affect severely the identification and tracking performance foroutdoor surveillance system. The video image pre-processing, to eliminate thevisual effects of images in a variety of weather conditions is essential for anall-weather outdoor vision systems. And rain is also the most frequent adverseweather conditions; the video image influenced by the rainfall will be severelydegraded. Therefore, detection and removal of raindrops in the video image is avery difficult task. This paper mainly studies the removal of rain in video image.Rain video is divided into two types of static scenes and dynamic scenes. Fordifferent situations study their solution, respectively.Raindrops remove in static scene. At present, the static algorithm cannotguarantee the effect of rain removal and real-time. So this paper presents thebackground subtraction based on sample model to detect of rain. Use the first frameof the video sequence to initialize the background, thus rapid, real-time backgroundsubtraction to detect raindrops. In addition, according to the intra-pixel spatialcorrelation, taking a simple method randomly selected eight neighborhood pixelvalues to create and update the sample model. When the removal of rain, we do notconsider the pixel information on a similar frame, but use the mean of the samplemodels instead of the brightness values of pixels covered by raindrop. Theexperiment proved that the algorithm not only has better real-time, but also caneffectively remove raindrops in the video image.Raindrops remove in dynamic scenes. As the problem that moving targetwill cause a lot of interference for rain removal. Study several ways of removingraindrop at present, and analysis of their advantages and disadvantages.Combination of the time domain statistical characteristics of raindrop, achieve analgorithm of rain removal under dynamic scene. Also use the temporal and spatialcorrelation between neighboring pixels to improve the traditional way of rainremoval. Different from previous method of rain removal, this algorithm is nolonger considered the shape,speed and other physical characteristics, colorcharacteristics, as well as the camera parameters, but according to the rata of change of the adjacent inter-frame brightness values of the raindrop, distinguishbetween raindrops and moving targets. What is more, on the basis of the traditionalmethod of rain removal, introduce neighborhood pixel information in the sameframe. Due to the adjacent inter-pixel has a very similar background, this methodeffectively achieve the removal of rain.