IP3 2021-Video & Image Processing 3 (LOT 12921)

This lot is generally related to image processing. Techniques are disclosed for modelling of dynamic scenes and detecting changes to efficiently identify foreground and background in moving pictures and videos. The technique inputs a current image and previously stored images, represents the current image in a subspace, learns a dynamical model, and adapts the subspace based on the scene changes. The techniques further adapt dynamical model parameters in accordance with the scene changes, determine difference between prediction and current image and outputs an indication of a new object. Few techniques compute optical flow for an image, perform an invariant transformation and estimate probability for a current input to belong to the background. The technology may be implemented in video modeling services, video processing applications, object detection, object tracking, object classification, event analysis, industrial automation, security and surveillance, real time video analysis, etc.