Binary Segmentation of Video Sequences in Real Time

We present a method for foreground-background video segmentation in real-time that may be used in applications as, for instance, Background Substitution, Analysis of Surveillance Cameras, Highway Cars Detection and so on. Our approach implements a probabilistic segmentation based on the binary Quadratic Markov Measure Fields models (QMMFs). That framework regularizes the likelihood of each pixel to belong to each one of the models (foreground and background). Then our proposal consists of a model for the likelihood that takes into account: an estimation of the static background, motion of the foreground, illumination changes and casted shadows. In order to fulfill the real-time requirement we implement a parallel version of our algorithm in CUDA using a NVIDIA GPU.