Real-time change detection for surveillance in public transportation

The efficiency of event detection in video surveillance applications depends on the performance of the early stage of content change detection. However, this stage encounters difficulties deriving from the presence of noise in image acquisition devices. Furthermore, illumination variations in the surveyed scene tend to cause degradation in change detection results. This paper introduces a real-time statistic change detection technique which employs a block-based clustering procedure for the estimation of the noise model. The method is applied in conjunction with a brightness normalization stage for the suppression of ambient illumination variations. Additionally, an adaptation of the technique for real-time surveillance of subway environments is proposed.