Image processing device and image processing method
An image processing device ( 100 ) includes an object detecting unit ( 103 a ) that detects objects each appearing in a corresponding one of pieces of image data; an object tracking unit ( 103 b ) that tracks the detected objects in a time direction; a descriptor generating unit ( 104 ) that generates feature descriptors of the respective objects on the basis of results of the detection and results of the tracking of the objects; a data recording control unit ( 105 ) that forms a database from the generated feature descriptors and the pieces of image data of the objects corresponding to the feature descriptors; an image retrieving unit ( 108 a ) that retrieves, from the database, object groups that match a retrieval condition and form a pair of object groups appearing in pieces of image data obtained by capturing different regions, each of the object groups corresponding to a corresponding one of the regions; a feature extracting unit ( 108 b ) that extracts a feature to be used for checking, from pieces of image data and feature descriptors of each of the retrieved object groups; and a determining unit ( 108 c ) that determines whether the object groups forming the pair of object groups are identical to each other, by checking the extracted features against each other.
1. An image processing device comprising:
processing circuitry
to detect objects each appearing in a corresponding one of pieces of image data, by analyzing the pieces of image data;
to track the objects detected, in a time direction;
to generate feature descriptors of the respective objects on a basis of results of the detection and results of the tracking;
to form a database from the feature descriptors generated and the pieces of image data of the objects corresponding to the feature descriptors;
to retrieve, from the database formed, object groups that match a set retrieval condition and form a pair of object groups appearing in pieces of image data obtained by capturing different regions, each of the object groups corresponding to a corresponding one of the regions;
to extract a feature to be used for checking, from pieces of image data and feature descriptors of each of the object groups retrieved; and
to determine whether the object groups forming the pair of object groups are identical to each other, by checking features against each other, each of the features being extracted, wherein the processing circuitry performs feature-based clustering between the pieces of image data of each of the retrieved object groups, using the pieces of image data and the feature descriptors of a corresponding one of the object groups retrieved, and extracts, from pieces of image data in a cluster defined by a result of the clustering, one feature for the cluster, the one feature being the feature to be used for checking.
2. The image processing device according to claim 1 , wherein each of the feature descriptors includes a visual feature of a corresponding one of the objects and location information of the corresponding one of the objects.
3. An image processing method comprising:
detecting objects each appearing in a corresponding one of pieces of image data by analyzing the pieces of image data;
tracking the detected objects in a time direction;
generating feature descriptors of the respective detected objects on a basis of results of the detection of the objects and results of the tracking of the detected objects;
forming a database from the generated feature descriptors and the pieces of image data of the objects corresponding to the feature descriptors;
retrieving, from the formed database, object groups that match a set retrieval condition and form a pair of object groups appearing in pieces of image data obtained by capturing different regions, each of the object groups corresponding to a corresponding one of the regions;
extracting a feature to be used for checking, from pieces of image data and feature descriptors of each of the retrieved object groups;
determining whether the object groups forming the pair of object groups are identical to each other, by checking features against each other, each of the features being extracted; and
performing feature-based clustering between the pieces of image data of each of the retrieved object groups, using the pieces of image data and the feature descriptors of a corresponding one of the retrieved object groups, and extracting, from pieces of image data in a cluster defined by a result of the clustering, one feature for the cluster, the one feature being the feature to be used for checking.