AU2017203772B2

AU2017203772B2 - A method of automatically calibrating an array of cameras, and a corresponding installation

Info

Publication number: AU2017203772B2
Application number: AU2017203772A
Authority: AU
Inventors: Christelle Maria France BAUDRY; Vincent Despiegel
Original assignee: Idemia Public Security France
Current assignee: Idemia Public Security France
Priority date: 2016-06-07
Filing date: 2017-06-05
Publication date: 2022-06-16
Anticipated expiration: 2037-06-05
Also published as: EP3255876B1; FR3052278B1; FR3052278A1; AU2017203772A1; EP3255876A1; EP3255876C0

Abstract

A B S T R A C T A method of automatically calibrating at least one first camera and at least one second camera that have 5 fields of view covering a location and presenting an overlap zone, and that are connected to a processor unit executing a program arranged to detect semantically defined characteristic points in an element, the method comprising the steps of: 10 - capturing respective images simultaneously with two cameras; • detecting in each image, in the overlap zone, semantically-defined characteristic points; and - putting two semantically-defined characteristic 15 points of the same type, each detected in one of the images, into correspondence in order to deduce therefore a relationship between the cameras. 1/1 50 1012 1023 51 52 53 101 102 103 1 3 Sole figure

Description

A B S T R A C T

A method of automatically calibrating at least one first camera and at least one second camera that have fields of view covering a location and presenting an overlap zone, and that are connected to a processor unit executing a program arranged to detect semantically defined characteristic points in an element, the method comprising the steps of: - capturing respective images simultaneously with two cameras; • detecting in each image, in the overlap zone, semantically-defined characteristic points; and - putting two semantically-defined characteristic points of the same type, each detected in one of the images, into correspondence in order to deduce therefore a relationship between the cameras.

1/1

50 1012 1023

51 52

53

101 102 103

1 3

Sole figure

A METHOD OF AUTOMATICALLY CALIBRATING AN ARRAY OF CAMERAS, AND A CORRESPONDING INSTALLATION The present invention relates to managing an array of cameras, e.g. suitable for use for biometric recognition purposes.

STATE OF THE PRIOR ART There exist installations having cameras with fields of view that cover a location while presenting an overlap zone and that are connected to a processor unit running a program for biometric face recognition. The program is arranged to detect a face in the images supplied thereto by the cameras. The program is also arranged to detect semantically-defined points in that face, such as: the right corner of the mouth; the left corner of the mouth; the outside corner of the left eye; the inside corner of the left eye; the outside corner of the right eye; the inside corner of the right eye; the tip of the nose; These characteristic points and others, are used for putting the image of the face into a reference position and subsequently for determining whether the face corresponds to a face having its biometric characteristics stored in a database or in an identity document. By way of example, the location in question is an entry airlock giving access to a secure enclosure to which access is authorized only to people recognized by the biometric recognition program. To improve the reliability of such recognition, it is important to be able to determine the position of the person in the environment, which can be done with a calibrated array of cameras. This calibration operation is presently performed by an operator when the installation is initially put into service, and periodically during maintenance operations. This operation is difficult, and if it is not repeated regularly, it cannot prevent drift appearing over time during the operation of the installation.

OBJECT OF THE INVENTION An object of the invention is to simplify the calibration of such cameras.

BRIEF SUMMARY OF THE INVENTION To this end, the invention provides a method of automatically calibrating at least one first camera and at least one second camera that have fields of view covering a location and presenting an overlap zone, and that are connected to a processor unit executing a program arranged to detect semantically-defined characteristic points in an element. The method comprises the steps of: • capturing respective images simultaneously with two cameras; • detecting in each image, in the overlap zone, semantically-defined characteristic points; and - putting two semantically-defined characteristic points of the same type, each detected in one of the images, into correspondence in order to deduce therefore a relationship between the cameras. Thus, setting up correspondence in the overlap zones makes it possible to create a relationship between the fields of view of the two cameras by associating a given point in three-dimensional space with the same semantically-defined point as detected in each of the images. According to one embodiment, the semantically defined characteristic points comprise at least one of the following points: - right corner of the mouth; - left corner of the mouth; - outside corner of the left eye; - outside corner of the right eye;

- inside corner of the left eye; - inside corner of the right eye; - inside center of the left eye; - inside center of the right eye; - tip of the nose. Other characteristics and advantages of the invention appear from reading the following description of particular, non-limiting embodiments of the invention.

BRIEF DESCRIPTION OF THE FIGURE Reference is made to the sole accompanying figure, which is a diagrammatic plan view of an installation for recognizing faces that enables the method of the invention to be performed.

DETAILED DESCRIPTION OF THE INVENTION With reference to the figure, the installation of the invention comprises at least a first camera 1, a second camera 2, and a third camera 3, which are connected to a processor unit 4 that is executing a program for biometric face recognition that is capable of detecting, in an image representing a face, semantically defined characteristic points and biometric data of the face. By way of example, the semantically-defined characteristic points may comprise: the right corner of the mouth; the left corner of the mouth; the outside corner of the left eye; the inside corner of the left eye; the outside corner of the right eye; the inside corner of the right eye; the tip of the nose. Such a program is itself known. By way of example, the biometric recognition program makes use of an algorithm of the scale-invariant feature transform (SIFT) type or of the speeded-up robust features (SURF) type. The cameras are positioned in a passage 50 so as to have fields of view 101, 102, and 103 that, in pairs, present overlap zones 1012 and 1023. More precisely, the field of view 101 covers the field of view 102 in the zone 1012; the field of view 102 covers the field of view 103 in the zone 1023; and the fields of view 101 and 103 do not overlap. The passage 50 is closed at both ends, respectively by an entry door 51 and by an exit door 52 so as to form an airlock. Opening of the entry door 51 is controlled by presenting and processing an identity document to a detector 53 connected to the control unit 4. The detector 53 is arranged to extract from the identity document biometric data of the face of the bearer of the identity document and to transmit that data to the processor unit 4. The data is contained either in a photograph integral with the identity document or else in an integrated circuit incorporated in the identity document. The exit door 52 is opened under the control of the processor unit 4 when the biometric data extracted by the processor unit 4 from the images provided by the cameras 1, 2, and 3 corresponds to the biometric data from the identity document. This nominal mode of operation is itself known and is not described in greater detail herein. In order to make the automatic recognition of faces in this nominal mode of operation perform better, it is necessary for the cameras 1, 2, and 3 to be calibrated. By way of example, such calibration enables the processor unit 4 to be sure that a face detected in an image provided by each camera 1, 2, and 3 is indeed to be found in the passage and not outside it, and to estimate the size of the image in order to verify that it matches theoretical anthropometric data. This enables the processor unit to detect an attempted fraud. In the method of the invention, this calibration is performed automatically by the processor unit 4 during a calibration operation initiated by the installer when putting the installation into service, and then during calibration operations that are initiated periodically in autonomous manner by the processor unit 4 during normal operation of the installation. Each calibration operation comprises the steps of: • capturing respective images simultaneously with two cameras; • detecting in each image of a pair, in the overlap zone, semantically-defined characteristic points; and - putting two semantically-defined characteristic points of the same type, each detected in one of the images of each pair, into correspondence in order to deduce therefore a relationship between the cameras. For example, the processor unit 4 analyzes two images captured at the same instant respectively by the camera 1 and by the camera 2 in order to detect therein semantically-defined characteristic points. In the image from the camera 1 there are detected: a tip of a nose, an outside corner of a right eye, and a right corner of a mouth. In the image from the camera 2, there are detected: a tip of a nose, an outside corner of a right eye, an inside corner of a right eye, an outside corner of a left eye, an inside corner of a left eye, a right corner of a mouth, and a left corner of a mouth. The processor unit 4 thus establishes correspondences between the semantically-defined characteristic points that are common to both images by comparing the positions of the tip of the nose, of the outside corner of the right eye, and of the right corner of the mouth in the image supplied by the camera 1 and in the image supplied by the camera 2. These correspondences are established by associating a point detected in each of the cameras (e.g. the corner of the right eye) with the same point in three-dimensional space. This makes it possible to create a relationship between the fields of view of the two cameras. After performing the method several times over, a step of analyzing the previously-established correspondences is performed in order to deduce therefrom a quality factor for the correspondences. Naturally, the invention is not limited to the implementations described, but covers any variant coming within the ambit of the invention as defined by the claims. In particular, the invention is not limited to applications for biometric face recognition. The invention is applicable to an installation that does not have biometric recognition means, e.g. an installation for video surveillance. It should be observed that it is also possible, in a variant of the invention, to deduce the relative positions of the cameras and to deduce therefrom treatment for application to the images in order to improve biometric performance (e.g. to straighten up an image when it is known that it has been taken as a low angle shot by a camera). The installation may be of a structure that is different and may have some other number of cameras, providing there are at least two cameras. In this specification, the terms "comprise", "comprises", "comprising" or similar terms are intended to mean a non-exclusive inclusion, such that a system, method or apparatus that comprises a list of elements does not include those elements solely, but may well include other elements not listed. The reference to any prior art in this specification is not, and should not be taken as, an acknowledgement or any form of suggestion that the prior art forms part of the common general knowledge.

Claims

CLAIMS 1. A method of automatically calibrating at least one first camera and at least one second camera that have fields of view covering a location and presenting an overlap zone, and that are connected to a processor unit executing a program arranged to detect semantically defined characteristic points in an element, the element being a face and the program being a program for biometric face recognition, wherein the semantically defined characteristic points comprise at least one of the following points: • right corner of the mouth; • left corner of the mouth; • outside corner of the left eye; outside corner of the right eye; • inside corner of the left eye; • inside corner of the right eye; • inside center of the left eye; • inside center of the right eye; tip of the nose, the method comprising the steps of: • capturing respective images simultaneously with two cameras; • detecting in each image, in the overlap zone, semantically-defined characteristic points; and - putting two semantically-defined characteristic points of the same type, each detected in one of the images, into correspondence in order to deduce therefore a relationship between the cameras.
2. A method according to claim 1, wherein, after performing the method several times over, a step is performed of analyzing the previously-established correspondences in order to deduce therefrom a quality factor for the correspondences.
3. A method according to claim 1, wherein the program is a biometric recognition program.