OA18310A

OA18310A - A system and a method for image recognition.

Info

Publication number: OA18310A
Application number: OA1201700265
Authority: OA
Inventors: Hongxin ZHANG; Mingxiu Chen; Ningqing Liang
Original assignee: Yutou Technology (Hangzhou) Co., Ltd.
Priority date: 2015-01-12
Filing date: 2015-06-12
Publication date: 2018-10-03

Abstract

The invention discloses an image recognition system, comprising: a robot having an image collection module configured to drive an image collection module to collect an image in a view of the robot by the image collection drive module; a light source configured to fill the light when the image collection module collecting an image; a face detection module configured to locate a face image in an image according to the image collected by the image collection module; a face recognition module configured to implement a preprocessing for the located face image, then the preprocessed face image being compared with an image feature information of a known identity in a database, to determine an identity information and a confidence rate of the present face image. The invention can identify a face based on a fixed face pose, and it can identify a face based on a local or web server database.

Description

1. Field of the Invention

The présent disclosure relates to a security field, more specifically, to a system and a method for image récognition for a robot system.

2. Descrintion of the Related Art

I

Nowadays, with the increasing requirements for the security, a great number of security Systems utilize password authentication for the identification. Notwithstanding, this kind of identification lacks in security, and people is easy to obtain the decryption method, thus the high-level security requirement is not satisfied. Therefore, the authentication mode through the identification of the fingerprint, iris and face is gradually adopted by the high-level security system. The human biological characteristics, such as face, fingerprint and iris, are innate, the uniqueness and the difficulty of being duplicated are necessary conditions for the identity authentication. The face récognition has peculiar features compared with other types of biological récognition, these feature are as follows:

non-obligatory: the user does not need to fit the face collection device, and the face collection device can obtain the face image when the

PauI JING râKl PcopERiy Attorney h! ACCREtJlTEd AÇENT user is unconscious, thus the collection method has no “mandatory”;

non-contact: the image is collected although the user does not need to directly contact the device;

concurrency: the sortîng, the détermination and the identification of a plurality of faces can be executed in the application scénarios;

visual characteristic: people are identified by their faces, and the operation is simple, the resuit is intuitional, the stealthiness is good.

At présent, the face récognition system generally comprises a caméra configured to collect the face image, a light source configured to fill light, an assistant position system or a sign configured to prompt the face collection location, a computer (such as an embedded computer) configured to execute the face récognition software, a device configured to process or display the identification resuit such as a reminder light, a relay configured to open the door, and a database sheet in order to record the identification results.

The face récognition system for the security system has the following issues in the application scénarios: 1. the requirement for the pose of the face collection is fixed; 2. the light condition is sensitive and should be fixed by filling the light; 3. it does not need high speed calculatîng cause the calculating demand is one-off.

² 35

SUMMARY OF THE INVENTION

To résolve the shortcoming of the prior art, the invention provides an image récognition system, comprises:

a robot having an image collection module, configured to drive the image collection module to collect an image in a view of the robot by the image collection drive module;

a light source configured to fill the light when the image collection module collecting an image;

a face détection module configured to locale a face image in the image collected by the image collection module;

a face récognition module configured to implement a preprocessing for the located face image, then the preprocessed face image being compared with an image feature information of a known identity contained in a database, to détermine an identity information and a confidence rate of the présent face image.

Preferably, the image collection module is a high définition caméra, the high définition caméra is capable to obtain static image and to collect at least 30 images per second;

the high définition caméra is connected to the robot by a MIPI or a USB interface.

Preferably, the light source comprises an ambient light source and an infrared light source;

upon utilizing the ambient light to fill the light and collecting the image by the image collection module, if the collected image can not be recognized, the light is filled by the infrared light source.

Preferably, the preprocessing comprising:

executing an angle correction treatment and a light treatment (such as brightness normalization and polarized light correction) for the face image presenting in the image.

Preferably, the database comprises a local data memory module and a web server data memory module.

Preferably, the robot further comprises a sounding device connected to the database, the sounding device sends out various types of prompt tones according to a comparative resuit from the face récognition module.

Preferably, the image récognition system further comprises a record feedback device to record and/or feed back a comparative resuit from the face récognition module.

Preferably, the face récognition module implements a comparison by SVM algorithm.

A method for the image récognition for the above-mentioned system, comprises the steps of:

(a) collecting the image in the view of the robot by the image collection module of the robot while filling the light by a light source as the image collection module collecting the image;

(b) using the face détection module to implement a location process to the face image presenting in the image which is collected by the image collection module;

(c) using the face détection module to implement a preprocessing for the face after being located, then comparing with an image feature information of a known identity contained in a database, to détermine an identity information and a confidence rate of the présent face image.

Preferably, in the above method, if the identity information of présent face image is incompatible with an image feature information of a known identity in a database, execute the step (a) to step (c) continually.

BRIEF DESCRIPTIONS OFTHE DRAWINGS

Referring to the accompanying drawings, the description made by the unlimited embodiments, the disclosure and the feature, outline and advantage thereof will be more obvious. The same signs in ail drawings indicate the same portions and they are not drawn in proportion intentionally. It illustrâtes the meaning of the invention.

Figure 1 shows the identity récognition system structure and the operation case thereof according to the présent invention.

DETAILED DESCRIPTIONS

The following description provides the details to permit a better compréhension of the invention. However, it is obvious for the people skilled in the art that the invention can be implemented without any one or more details. In other examples, to avoid confusion, the known technical features in the art are not described.

To understand the invention thoroughly, the following descriptions will provide detail steps and structures to explain the technical solution for the invention. The preferred embodiment is described as follows. However, the invention has further embodiments beyond the detailed description.

Since the invention consists in a part of the robot Visual system to implement the face identification, the issues in the application of the robot should be addressed. The issues comprise: 1 various poses of the face presenting in a view of the robot; 2. various light condition, including polarized light or no filling light source; 3. real-time recognizing the face in the view of the robot that demanding a quîck response, and the real-time feedback implemented through a continuous récognition in accordance with the variation of the identified face.

To résolve the above issues, the embodiment provides an image récognition system, comprising:

a face détection module configured to locate a face image in the image collected by the image collection module;

In an embodiment of the invention, it is optional but unlïmited that, the image collection module of the robot is a high définition caméra. Preferably, the high définition caméra is capable to obtain static image and to collect at least 30 images per second, tofurther meet the requirement of high-speed image collection. For example, even if the object in the view of the robot moves fast, the invention is also able to collect the image clearly. It is optional but unlimited that, the high définition caméra is connected to the robot by a MIPI or an USB interface. In some optional embodiments, the robot can implement a real-time adjustment for the framing scope or angle of the image collection module througha motor. For example, if the image collection module detects people crossing within the visual scope of it, the image collection module can implement real-time trackîng snapshot by the motor, such as moving with the object simultaneously and implementing the enlarged amplifying snapshot immediately, to improve the définition of the collected image.

In an embodiment of the invention, it is optional but unlimited that, the above light source includes an ambient light source and an infrared light source. The ambient light source is a build-in light source of the robot. The advantage of the ambient light source is homogeneity of the lighting. However the disadvantage of the ambient light source is that the luminance is low and it can not be directed. Furthermore, the luminance can be controlled by other high-level application, the light even be closed sometimes. Hence, the ambient light can not meet the requirement of filling light. Because a set of infrared lîght-emitting device is added to the invention and is used to fill the light of the image, the lumination power is controlled by the image récognition system limitedly, to achieve stable filling light in various scénarios. For example, when the ambient light source is used to fill light and to collect image by the image collection module, if the collected image can not meet the requirement of the récognition, then the invention uses the infrared light source to fill the light, thus to obtain a clear image.

In an embodiment of the invention, it is optional but unlimited that, the face détection module implements the location, that is, the face position is located in the fùll-fïeld image collected by the robot. In the présent security system, this step is unnecessary as the position is fixed. Then preprocess the located face image by the face récognition module, the preprocessed face image is compared with an image feature information of a known identity in a database, to détermine an identity information and a confidence rate of the présent face image. The face récognition module can execute the angle correction treatment and the light treatment (such as brightness normalization and polarized light correction) for the face image presented in the image. Due to the great change of the ambient and angle of the face collection, the invention increases the récognition rate through the execution of the angle correction treatment for the face image collected and located by the face détection module. Meanwhile, the embedded récognition technology also can process the light of the image, to facilitate the comparison and to increase the accuracy.

In an embodiment of the invention, it is optional but unlimited that, the above database includes a local data memory module and a web server data memory module. Based on the embedded system of the robot, the local data memory module adopts the face récognition technology according to the feature matching. Firstly, the local data memory module makes the feature database of the face image of a known identity, secondly it extracts a same type of feature according to the face image of the real-time collection, comparing the feature distance of the présent face to the feature distance of the database face through a math distance function. Finally, it détermines the possible identity, and gives out the confidence rate. Since the web server data memory module has more computing resources and more flexible application architectures, it adopts the face récognition technology based on a deep leaming model which is configured to generate the face feature in the database, to construct the face category în the database by SVM (Support Vector Machine) algorithm or other standard classifiers. Then it calculâtes the model feature according to the face image of the real-time collection, and détermines the identity and confidence rate of the face image by the classifier. The récognition technology of the embedded feature matching supports the identification of 20-50 people. In a certain variation range of light and angle, the récognition accuracy rate of 20 people is greater than 90%, the récognition accuracy rate of 50 people is greater than 80%, The récognition technology on the server based on the deep leaming supports the récognition from 50 people to hundreds people at least, and the récognition accuracy rate is greater than 97%.

In an embodiment of the invention, it is optional but unlimited that, the robot has a sounding device connected to the above database, the sounding device sends out various types of prompt tones according to a comparative resuit from the face récognition module. For example, if to the face récognition module has a correct comparison, the sounding device gets the prompt tones corresponding to the présent face image in the database, such as “Hello, Mr. Chen”. If the the regarded identification has failed after crosschecks, the greeting application still can sends out a general greeting without the identify information, such as sending out a simple “Hello” by the sounding device. In an optional embodîmenL the invention can be connected to the access control system. If the identification succeeds, the access is allowed, otherwise if the identification fails, the access is not allowed.

In an embodiment of the invention, it is optional but unlimited that, the image récognition system provided by the invention further comprises a record feedback device to record and/or feed back a comparative resuit from the face récognition module. It is an optional member, the record function and the feedback function don’t need to be implemented simultaneously. In some scénarios, only one of them is needed.

Meanwhile, the invention also provides a récognition method by the above-mentioned image récognition system, comprising the steps of:

(a) collecting the image in the view of the robot by the image collection module of the robot, and filling the light through a light source as the image collection module collecting the image;

(b) using the face détection module to implement a location n

process to the face image presentîng in the image which is collected by the image collection module;

(c) using the face détection module to implement a preprocessing for the face image after being located, then comparing with an image feature information of a known identity contained in a database, to détermine an identity information and a confidence rate of the présent face image.

If the identity information of présent face image is incompatible with an image feature information of a known identity in a database, execute the step (a) to step (c) continuai ly.

Figure 1 illustrâtes completely the operation process of the identity récognition system: firstly, the greeting application of the robot sends a récognition request to the image collection drive module, the image collection drive module accepts the request, and transmits the image sent from the caméra to the face détection module. The détection software truncates the located face image and executes the preprocess, and then sends the processed image to the face récognition module. The face récognition module transmits the récognition resuit to the resuit vérification module, the system executes the corresponding operation according to the determining resuit. If the resuit is correct, the resuit is sent to the greeting application, and the application utilizes the sounding device to greet the user in the image of the caméra according to the recognized identity; if the resuit is wrong, the system retransmits the récognition request to the image collection drive, and reenters into the récognition procedure. The main point of the détermination of the récognition resuit vérification module dépends on the confidence rate of the transmitting resuit of the face récognition software. In the worst case, when the identification fails, the times of retransmîtting request is controlled by the greeting application according to a request time out, to détermine whether the request has to be retransmitted. If the correct resuit is not obtained after the identification is time out, the identification is failed. Then the greeting application still can sends out a general greeting without identify information, such as a simple “Hello”.

Therefore, as the invention adopts the above technical solution, the invention does not need fixed the face pose to identify a face. Meanwhile, the invention can identify a face based on a local or web server database, and it increases the accuracy of the identification. Furthermore, the face récognition procedure meets the real-time requirement through the suitable caméra, the operation hardware module and the operation frame.

While the présent disclosure has been described in connection with certain exemplary embodiments, it is to be understood that the invention is not limited to the disclosed embodiments, the device and structure, which are not specifically described, should be understood as the common manner in the art to be implemented; any people skilled in the art can make possible changes and modifications, or équivalents thereof for the technical solution of the invention according to the above method without falling out of the scope of the invention. Therefore, the various modifications and équivalent arrangements without departing away from the technical solution of the invention, are included within the spirit and the scope of the technical solution of the invention.

Claims

What is claimed is:

1. An image récognition system, comprising:

a robot, having an image collection module, configured to drive the image collection module to collect an image in a view of the robot by an image collection drive module;

a light source configured to fill the light when the image collection module collecting an image;

a face détection module configured to locate a face image in the image collected by the image collection module;

a face récognition module configured to implement a preprocessing for the located face image, then the preprocessed face image being compared with an image feature information of a known identity contained in a database, to détermine an identity information and a confidence rate of the présent face image.
2. The image récognition system as claimed in Claim 1, wherein the image collection module is a high définition caméra, the high définition caméra îs capable to obtain static image and to collect at least 30 images per second;

the high définition caméra is connected to the robot by a MIPI or a USB interface.

.__. j'NG < > b /-Hbfr!gaÆ'.l PropEiny AnouwEy

S*' fri—G— ΟΆΡ1 AccRcdtrcdaçent
3. The image récognition System as claimed in Claim l, wherein the light source comprises an ambient light source and an infrared light source;

upon utilizing the ambient light to fill the light and collecting the image by the image collection module, if the collected image can not be recognized, the light is filled by the infrared light source.
4. The image récognition system as claimed in Claim 1, the preprocessing comprising:

executing an angle correction treatment and a light treatment for the face image presenting in the image.
5. The image récognition system as claimed in Claim 1, wherein the database comprises a local data memory module and a web server data memory module.
6. The image récognition system as claimed in Claim 1, wherein the robot further comprises a soundîng device connected to the database, the sounding device sends out various types of prompt tones according to a comparative resuit from the face récognition module.
7. The image récognition system as claimed in Claim l, further comprising a record feedback device to record and/or feed back a comparative resuit from the face récognition module.
8. The image récognition system as claimed in Claim 1, wherein the face récognition module implements a comparison by SVM algorithm.
9. An image récognition method, using an image récognition system as claimed in any one of Claims 1-8, comprising the steps of:

(a) collecting an image in the view of the robot by the image collection module of the robot while filling the light by a light source;

(b) using the face détection module to implement a location process to the face image presenting in the image which is collected by the image collection module;

(c) using the face détection module to implement a preprocessing for the face image after being located, then comparing with an image feature information of a known identity contained in a database, to détermine an identity information and a confidence rate of the présent face image.
10. The image récognition method as claimed in Claim 9, wherein in the step (c), if the identity information of the présent face image is incompatible with the image feature information of a known identity in a database, execute the step (a) to step (c) continualiy.