Locally adaptive thresholding of the sequence of image frames

A technique for real-time segmentation of image frames of a dynamic scene having spatial and temporal variable illumination is described. The technique combines locally adaptive thresholding and a fast region growing algorithm. The thresholding algorithm is both spatially and temporally adaptive. It is based on local thresholds organised in a threshold map so that each element of the map represents the threshold for a single square image sector. During processing, elements of the map are dynamically being updated in order to adapt the map to the perceived local illumination changes over time. The updating is driven by a priori knowledge about the dynamic scene and the knowledge accumulated during past processing.