Method for identifying a person in a video, by a visual signature from that person, associated computer program and device

1. A method for identifying a person in a video, by a visual signature from said person, the method comprising:

for each image of a plurality of successive images of a video stream from a camera, searching for at least one person present in the each image and defining, in the each image, for each person found, a field, called person field, at least partially surrounding said each person found;

for each person found of at least one person found, gathering into a track segment several person fields derived from successive images of said plurality of successive images and at least partially surrounding said each person found;

for each track segment, identifying the each person found in the track segment by a visual signature from the each person found, said identifying comprising:

for each person field in the track segment, determining said visual signature from the each person found in that track segment, called local visual signatures,

determining an aggregated visual signature from the local visual signatures, and

identifying the each person found in the track segment from the aggregated visual signature;

for each track segment,

for each person field in the track segment, searching for at least one number present in the each person field and defining, in the each person field, for each number found, a field, called number field, surrounding the each number found;

for each number field in the track segment, recognizing the each number found present in the each number field as recognized numbers and, for each number recognized, evaluating a reliability of the recognizing;

selecting one number of the recognized numbers from the reliability of the recognized numbers; and

searching for the one number selected from a set of predefined identification numbers identifying respective persons;

wherein the identifying of the each person found in said track segment by said visual signature from the each person found is carried out if the each person found could not be identified by an identification number.

2. The method of claim 1 , wherein the aggregated visual signature is a mean of the local visual signatures of the person.

3. The method of claim 1 further comprising, for each determination of a local visual signature, evaluating a reliability of that local visual signature and wherein the aggregated visual signature is determined from, additionally to the local visual signatures, their associated reliability.

4. The method of claim 1 , wherein said selecting one number of the recognized numbers from the reliability associated with said recognized numbers comprises:

filtering each number whose reliability is less than a predefined threshold; and

selecting one other number, called reliable numbers, from their associated reliability.

5. The method of claim 4 , wherein said selecting one other number of the reliable numbers from their associated reliability comprises selecting, from among values of the reliable numbers, one of which a combination, such as a sum or a mean, of the reliability of the numbers having that value is highest, and wherein the one other number selected is that having that value.

6. The method of claim 1 , wherein identifying the person in the track segment from the aggregated visual signature comprises:

determining, among the predefined identification numbers, a predefined identification number associated with one or more reference visual signatures having a distance from the aggregated visual signature less than a predefined threshold.

7. The method of claim 1 , further comprising, for each track segment:

determining, among the person fields in the track segment, which one person field first crosses, in a predefined direction, a line having a fixed and predefined position in the plurality of successive images; and

determining a crossing instant of crossing the line from an instant associated with an image containing the one person field crossing the line first.

8. A non-transitory computer readable medium, comprising a program including instructions for the executing steps of a method for identifying a person in a video, by a visual signature from said person, when said program is executed on a computer, the method comprising:

for each image of a plurality of successive images of a video stream from a camera, searching for at least one person present in the each image and defining, in the each image, for each person found, a field, called person field, at least partially surrounding said each person found;

for each person found of at least one person found, gathering into a track segment several person fields derived from successive images of said plurality of successive images and at least partially surrounding said each person found;

for each track segment, identifying the each person found in the track segment by a visual signature from the each person found, said identifying comprising:

for each person field in the track segment, determining said visual signature from the each person found in that track segment, called local visual signatures,

determining an aggregated visual signature from the local visual signatures, and

identifying the each person found in the track segment from the aggregated visual signature;

for each track segment,

for each person field in the track segment, searching for at least one number present in the each person field and defining, in the each person field, for each number found, a field, called number field, surrounding the each number found;

for each number field in the track segment, recognizing the each number found present in the each number field as recognized numbers and, for each number recognized, evaluating a reliability of the recognizing;

selecting one number of the recognized numbers from the reliability of the recognized numbers; and

searching for the one number selected from a set of predefined identification numbers identifying respective persons;

wherein the identifying of the each person found in said track segment by said visual signature from the each person found is carried out if the each person found could not be identified by an identification number.

9. A device for identifying a person in a video, by a visual signature from said person, comprising:

means designed, for each image of a plurality of successive images of a video stream from a camera, for searching for at least one person present in the each image and defining, in the each image, for each person found, a field, called person field, at least partially surrounding said each person found;

means designed, for each person found of at least one person found, for gathering into a track segment several person fields derived from successive images of said plurality of successive images and at least partially surrounding said each person found;

means designed, for each track segment, for identifying the each person found in the track segment by a visual signature from the each person found, said means for identifying comprising:

means designed, for each person field in the track segment, for determining a visual signature of the each person found in the person field in the track segment, called local visual signatures;

means designed for determining an aggregated visual signature from the local visual signatures; and

means designed for identifying the each person found in that track segment from the aggregated visual signature;

wherein for each track segment,

for each person field in the track segment, means designed for searching for at least one number present in the each person field and defining, in the each person field, for each number found, a field, called number field, surrounding the each number found;

for each number field in the track segment, means designed for recognizing the each number found present in the each number field as recognized numbers and, for each number recognized, evaluating a reliability of the recognizing;

means designed for selecting one number of the recognized numbers from the reliability of the recognized numbers; and

means designed for searching for the one number selected from a set of predefined identification numbers identifying respective persons;

wherein the identifying of the each person found in said track segment by said visual signature from the each person found is carried out if the each person found could not be identified by an identification number.