To realize a system in which excessive detections and undetections are few by calculating a plurality of feature variables of each small area from an image of a far infrared wavelength region and an image of a visible wavelength region, deciding reference with which the multi-dimensional space of the feature variables is classified into several areas and classifying the feature variable space into several areas according to the reference.
An image picking up part 11 sends the luminance values of respective wavelengths of RGB of a visible camera at each photographing place and the luminance value of an infrared camera. A feature variable calculating part 12 calculates a plurality of feature variables from the plurality of luminance values. A classification reference deciding part 13 decides classification reference. As a classification method, first, the centers of a fixed number of clusters are arranged at an appropriate places in the feature variable space as initial cluster positions, and the cluster centers are moved to divide the distribution of feature variable space by repeatedly performing an operation. A classification part 14 classifies individual pixels based on the position of the cluster centers decided by the part 13.
IMAGAWA TARO
MEGATA TSUYOSHI
Next Patent: SYSTEM FOR PROVIDING TRAFFIC INFORMATION