Information processing apparatus, method of controlling the same, and storage medium

An information processing apparatus which extracts a feature t of a person in frames received from at least one camera and registers the feature in a collation database so as to search for a person inputted as a query by referring to the collation database, the information processing apparatus comprising:

• one or more processors;
• and at least one memory coupled to the one or more processors, the at least one memory storing instructions which, when executed by the one or more processors, cause the information processing apparatus to:
• receive a frame sequentially from the camera as a video;
• detect a person from the frame;
• if a new person is detected, set the new person as a target to be tracked;
• track, in the frames following the frame, the person set as the target to be tracked until the person is not detected;
• extract a feature of the person set as the target to be tracked;
• and select, every time a predetermined period has elapsed even if the tracking of the person was not over, a predetermined number of features among features extracted from frames received during the predetermined period;
• and register the selected features in the collation database, wherein the person set as the target to be tracked is detected after a first period, in the collation database, the predetermined number of features among features extracted from frames received during a second period after the end of the first period is registered if the tracking of the person is continued, and wherein the first period is shorter than the second period and wherein the features are registered by causing the information processing apparatus to:
• classify faces of the tracked persons into a plurality of predetermined groups on the basis of each of directions of a frontal face, where the plurality of predetermined groups includes a group corresponding to a direction, as a reference direction, in which the frontal face is toward the camera and groups corresponding to directions different to the reference direction;
• and select, from the classified groups, faces in a number not more than a first predetermined number, and wherein, in the collation database, the features of the selected faces are registered.

A method of controlling an information processing apparatus which extracts a feature of a person in frames received from at least one camera and registers the feature in a collation database so as to search for a person inputted as a query by referring to the collation database, comprising:

• receiving, via a communication interface for communicating with at least one camera, a frame sequentially from the camera as a current frame of a live video;
• detecting a person from the received current frame;
• setting, if a new person is detected in the detecting, the new person as a target to be tracked;
• tracking, in the frames following the current frame, the person set as the target to be tracked until the person is not detected;
• extracting a feature of the person from each frame during the tracking of the person;
• selecting, every time a predetermined period has elapsed even if the tracking of the person was not over, a predetermined number of features among the features extracted from frames received during the predetermined period;
• and registering the selected features in the collation database, wherein the person set as the target to be tracked is detected after a first period, in the collation database, the predetermined number of features among features extracted from frames received during a second period after the end of the first period is registered if the tracking of the person is continued, and wherein the first period is shorter than the second period and wherein the features are registered by:
• classifying faces of the tracked persons into a plurality of predetermined groups on the basis of each of directions of a frontal face, where the plurality of predetermined groups include a group corresponding to a direction, as a reference direction, in which the frontal face is toward the camera and groups corresponding to directions different to the reference direction;
• and selecting, from the classified groups, faces in a number not more than the predetermined number, and wherein, in the collation database, the features of the selected faces are registered.

A non-transitory computer-readable storage medium storing a computer program which causes a computer to executes steps of a method of controlling an information processing apparatus which extracts a feature of a person in frames received from at least one camera and registers the feature in a collation database so as to search for a person inputted as a query by referring to the collation database, the method comprising:

• receiving, via a communication interface for communicating with at least one camera, a frame sequentially from the camera as a current frame of a live video;
• detecting a person from the received current frame;
• setting, if a new person is detected in the detecting, the new person as a target to be tracked;
• tracking, in the frames following the current frame, the person set as the target to be tracked until the person is not detected;
• extracting a feature of the person from each frame during the tracking of the person;
• selecting, every time a predetermined period has elapsed even if the tracking of the person was not over, a predetermined number of features among the features extracted from frames received during the predetermined period and registering the selected features in the collation database, wherein the person set as the target to be tracked is detected after a first period, in the collation database, the predetermined number of features among features extracted from frames received during a second period after the end of the first period is registered if the tracking of the person is continued, and wherein the first period is shorter than the second period and wherein the features are registered by:
• classifying faces of the tracked persons into a plurality of predetermined groups on the basis of each of directions of the faces a frontal face, where the plurality of predetermined groups include a group corresponding to a direction, as a reference direction, in which the frontal face is toward the camera and groups corresponding to directions different to the reference direction;
• and selecting, from the classified groups, faces in a number not more than the predetermined number, and wherein, in the collation database, the f features of the selected faces are registered.

A method of controlling an information processing apparatus which extracts a feature of a person in frames received from at least one camera and registers the feature in a collation database so as to search for a person inputted as a query by referring to the collation database, comprising:

• receiving a frame sequentially from the camera as a video;
• detecting a person from the frame;
• setting, if a new person is detected in the detecting, the new person as a target to be tracked;
• tracking, in the frames following the frame, the person set as the target to be tracked until the person is not detected;
• extracting a feature of the person set as the target to be tracked;
• and selecting, every time a predetermined period has elapsed even if the tracking of the person was not over, a predetermined number of features among features extracted from frames received during the predetermined period and registering the selected features in the collation database, wherein the person set as the target to be tracked is detected after a first period, in the collation database, the predetermined number of features among features extracted from frames received during a second period after the end of the first period is registered if the tracking of the person is continued, and wherein the first period is shorter than the second period and wherein the features are registered by:
• classifying faces of the tracked persons into a plurality of predetermined groups on the basis of each of directions of the faces a frontal face, where the plurality of predetermined groups includes a group corresponding to a direction, as a reference direction, in which the frontal face is toward the camera and groups corresponding to directions different to the reference direction;
• and selecting, from the classified groups, faces in a number not more than a first predetermined number, and wherein, in the collation database, the features of the selected faces are registered.

A non-transitory computer-readable storage medium storing a computer program which causes a computer to executes steps of a method of controlling an information processing apparatus which extracts a feature of a person in frames received from at least one camera and registers the feature in a collation database so as to search for a person inputted as a query by referring to the collation database, the method comprising:

• receiving a frame sequentially from the camera as a video;
• detecting a person from the frame;
• setting, if a new person is detected in the detecting, the new person as a target to be tracked;
• tracking, in the frames following the frame, the person set as the target to be tracked until the person is not detected;
• extracting a feature of the person set as the target to be tracked;
• and selecting, every time a predetermined period has elapsed even if the tracking of the person was not over, a predetermined number of features among features extracted from frames received during the predetermined period and registering the selected features in the collation database, wherein the person set as the target to be tracked is detected after a first period, in the collation database, the predetermined number of features among features extracted from frames received during a second period after the end of the first period is registered if the tracking of the person is continued, and wherein the first period is shorter than the second period and wherein the features are registered by:
• classifying faces of the tracked persons into a plurality of predetermined groups on the basis of each of directions of the faces a frontal face, where the plurality of predetermined groups includes a group corresponding to a direction, as a reference direction, in which the frontal face is toward the camera and groups corresponding to directions different to the reference direction;
• and selecting, from the classified groups, faces in a number not more than a first predetermined number, and wherein, in the collation database, the features of the selected faces are registered.