| 摘要 |
The present invention provides a video search system in which the repetitive work of condition setting and search execution is automated. A video search server of this system records videos from a plurality of cameras installed in an urban district, designates an object in some of the videos, and executes a search to determine whether the object is seen in others of the videos. Whether one and other objects are identical is determined by the degree of similarity in an object feature including the feature value of an image, etc., and, when the objects are determined to be identical, the object feature is updated or added. Accuracy is thereby improved because the updated object feature is used in subsequent search processes. A movement path is estimated by linking the search results of the identical objects together. In addition to movement speed, direction of advance, etc., when the object passes through a visual field of the camera, the object feature includes a category for distinguishing the object as a person, an automotive vehicle, a bicycle, etc., and information useful for identifying the object for each category (the feature value of a face, in the case of a person). When a search is executed from videos from the plurality of cameras, a camera in a high-probability path is prioritized, on the basis of an already discovered location and the direction of advance to that location, as the search is processed. |
