Foreground detector for video analytics system

摘要: Techniques are disclosed for creating a background model of scene using both pixel based approach and context approach. The combined provides an effective technique segmenting foreground from in frames video stream. Further, this can scale to process large numbers camera feeds simultaneously, e.g., parallel processing architectures, while still generating accurate model. ensures that the analytics system effectively efficiently respond changes scene, without overly increasing computational complexity. In addition, techniques updating model, frame-to-frame, by absorbing pixels into via absorption window, dynamically background/foreground thresholds.