Apparatus and method for inferencing topology of multiple cameras network by tracking movement

A method of tracking movements of objects to infer a topology of a network of multiple cameras, comprising:

• sequentially obtaining images from the multiple cameras;
• extracting at least one moving object from each of the obtained images;
• determining and registering appearing cameras and appearing times at which the moving objects appear, and disappearing cameras and disappearing times at which the moving objects disappear, for the multiple cameras;
• and
• inferring the topology of the network using the appearing cameras and appearing times and the disappearing cameras and disappearing times,
• wherein the inference of the topology comprises:
• if the moving objects appear on the cameras, determining whether the moving objects are new moving objects;
• if the moving objects are the new moving objects, registering the moving objects and generating image nodes of the appearing cameras, and if the moving objects are registered moving objects, comparing moving times of the moving objects with a threshold moving time;
• and
• if the moving times are greater than the threshold moving time, re-registering the moving objects as new moving objects and generating image nodes of the appearing cameras, and if the moving times are not greater than the threshold moving time, comparing the appearing cameras with previous disappearing cameras to determine the topology of the network,
• the method further comprising:
• if the moving objects disappear from the cameras, determining whether the moving objects appear on other cameras except the disappearing cameras;
• if the moving objects do not appear on the other cameras, generating non-overlapping nodes;
• and
• if the moving objects appear on the other cameras, determining whether previous appearing cameras are the same as the disappearing cameras to generate camera nodes or regression visible edges,
• wherein the generation of the camera nodes or the regression visible edges comprises:
• if the previous appearing cameras are not the same as the disappearing cameras, generating image nodes of the disappearing cameras;
• and
• if the previous appearing cameras are the same as the disappearing cameras, generating regression visible edges that regress to the disappearing cameras,
• wherein the generation of the image nodes of the disappearing cameras comprises generating overlapping edges to the image nodes, wherein weights of the overlapping edges to the image nodes and weights at the regression visible edges are the moving times.

A method of tracking movements of objects to infer a topology of a network of multiple cameras, comprising:

• sequentially obtaining images from the multiple cameras;
• extracting at least one moving object from each of the obtained images;
• determining and registering appearing cameras and appearing times at which the moving objects appear, and disappearing cameras and disappearing times at which the moving objects disappear, for the multiple cameras;
• and
• inferring the topology of the network using the appearing cameras and appearing times and the disappearing cameras and disappearing times,
• wherein the inference of the topology comprises:
• if the moving objects appear on the cameras, determining whether the moving objects are new moving objects;
• if the moving objects are the new moving objects, registering the moving objects and generating image nodes of the appearing cameras, and if the moving objects are registered moving objects, comparing moving times of the moving objects with a threshold moving time;
• and
• if the moving times are greater than the threshold moving time, re-registering the moving objects as new moving objects and generating image nodes of the appearing cameras, and if the moving times are not greater than the threshold moving time, comparing the appearing cameras with previous disappearing cameras to determine the topology of the network,
• wherein the determination of the topology of the network comprises:
• if the previous disappearing cameras are not registered or the appearing cameras of the moving objects are not the same as the previous disappearing cameras, determining whether the moving objects disappear from other cameras to generate image nodes or overlapping nodes;
• and
• if the appearing cameras of the moving objects are the same as the previous disappearing cameras, generating virtual nodes of the appearing cameras,
• wherein the generation of the image nodes or the overlapping nodes comprises:
• if the moving objects do not appear on the other cameras, generating the image nodes of the appearing cameras;
• and
• if the moving objects appear on the other cameras, generating overlapping nodes between the appearing cameras and the other cameras,
• wherein the generation of the image nodes comprises:
• generating invisible edges to the image nodes;
• and
• the generation of the overlapping nodes comprises generating visible edges to the overlapping nodes, wherein weights of the invisible edges to the image nodes and weights of the visible edges to the overlapping nodes are the moving times.