Research On Rain Removal In Video

Computer vision system has been widely applied in the field of industry, military and scientific research. However, bad weather brings enormous difficulties for detection and recognition in vision system. To solve this problem, this paper presents the research work under the rain and snow, especially the removal of rain in the video.Traditional methods usually remove rain in R, G and B components, respectively. This study found that only Y component is affected by raindrops after video transformed from the RGB to YCbCr color space. Based on this characteristic, this paper proposes a method that removing rain in Y component based on color space transformation. Experiments prove that our algorithm is time-saving and effective to recover non-rain scenes in the same level as the method processed in three components.Based on the raindrops motion characteristics, a novel rain detection method that relies on independent components analysis technology is proposed. Firstly, this study improves the moving objects detection algorithm based on FastICA, and then uses this into rain detection. Improved algorithm is fast and accurate for extracting raindrops, which lays foundation for the following work.The identification of rain and non-rain is the important part of rain removal algorithm. The criterion condition is single and the accuracy is not high in existing raindrops identification method. According to this problem, this paper presents a new method combining area, direction and width statistical information to identify raindrops. Experimental results show that proposed method is more accurate for raindrops identification, and the probability of false identification is relatively low.For the effect of using a few frames to remove rain is not ideal, two-step method is proposed to solve this problem. Firstly, removing raindrops in a single frame, and then combining three frames to remove rain further. Experiments show that the two-step method to remove rain has the similar result with using large amounts of frames. Meanwhile, the proposed algorithm is robust to complex scenes and different rainfall condition.