Real-time precipitation suppression in video streams
Adam N. Nilsson, Hoa Truong · Lund University Publications Student Papers (Lund University) · 2014
In surveillance cameras rain and snow can introduce an unwelcome noise to the video stream. The resulting effect of the rain becomes bright streaks in the frames of the video. These streaks can disturb human viewers and image processing algorithms. Rain streaks can be hard to detect and remove as they are a very dynamic phenomenon dependent on camera settings and weather conditions. This thesis aims to research some already invented rain removal algorithms and compare and evaluate them. Surveillance cameras supply video in real-time so it is not possible to access the whole video and perform heavy computations relying on information from the future. Qualities such as level of streak suppression and time required to perform necessary calculations are weighed against each other.