Method and system for automated face detection and recognition
원문보기
IPC분류정보
국가/구분
United States(US) Patent
등록
국제특허분류(IPC7판)
G06K-009/00
H04N-007/18
G06T-007/00
G06K-009/62
출원번호
US-0971008
(2008-01-08)
등록번호
US-9367730
(2016-06-14)
우선권정보
RU-2007102021 (2007-01-09)
발명자
/ 주소
Irmatov, Anwar Adkhamovich
Bazanov, Peter Valerievich
Buryak, Dmitry Yurievich
Kuznetsov, Victor Dmitrievich
Mun, Wang-Jin
Yang, Hae-Kwang
Lee, Yong-Jin
출원인 / 주소
S1 Corporation
대리인 / 주소
Schwegman Lundberg & Woessner, P.A.
인용정보
피인용 횟수 :
0인용 특허 :
2
초록▼
The present invention relates to a figure recognition system and method for automatic detection, tracking and recognition of a human face image. 2D image data in the surveillance zone are remotely collected by using an optical sensor, the faces of all persons in the surveillance zone are detected, a
The present invention relates to a figure recognition system and method for automatic detection, tracking and recognition of a human face image. 2D image data in the surveillance zone are remotely collected by using an optical sensor, the faces of all persons in the surveillance zone are detected, and corresponding positions are determined. The face is detected, the detected face's feature coordinate is estimated, and the detected face and the feature are tracked in the next frame while processing the video sequence. Image quality of each detected face is determined according to parameters of focus, brightness, contrast, and the presence of glasses. Recognition methods stored in the repository for each detected face are adjusted by considering the face image quality computation value, and a biometric feature set is generated by using the recognition method selected for each detected face. The figure is recognized according to the watch list by using the biometric feature generated by comparing each detected face and a template set stored in the database. A new user registration process is performed and the recognition method is adapted automatically by considering the watch list.
대표청구항▼
1. A figure recognition method in a method for a biometric figure recognition system with a watch list to automatically detect, track, and recognize a face image of a figure, the method comprising: a) remotely collecting 2D data in a monitoring area by using an optical sensor;b) detecting faces and
1. A figure recognition method in a method for a biometric figure recognition system with a watch list to automatically detect, track, and recognize a face image of a figure, the method comprising: a) remotely collecting 2D data in a monitoring area by using an optical sensor;b) detecting faces and face features of persons in the monitoring area, and determining positions thereof;c) finding a person and estimating the person's features coordinate;d) tracking the detected faces and face features in a subsequent frame during a video sequence;e) estimating image quality of each detected person according to parameters of focus, brightness, contrast, and glasses wearing state;f) controlling recognition methods stored in a repository by considering face image quality values for the respective detected persons;g) forming biometric features of each detected person by using a selected recognition method;h) comparing each detected person and a template set stored in a database by using the formed biometric features to identify a figure according to a watch list; andi) registering a new user, and performing an automatic recognition method adaptation process by considering the watch list. 2. The method of claim 1, wherein the step of i) further comprises selecting a best preprocessing method by using template bases of registered users, and controlling various measurement combination methods of face image similarity, when performing an automatic recognition method adaptation process in consideration of the watch list. 3. The method of claim 2, wherein the method further comprises selecting a Gabor optical filter by applying an AdaBoost automatic process when selecting the best preprocessing method. 4. The method of claim 2, wherein the controlling of face image similarity measurement combination methods includes automatic generation of a new classification rule based on a SVM (Support Vector Machine), and calculation of a measured combination coefficient using the AdaBoost process. 5. The method of claim 1, wherein the controlling recognition methods is performed by consideration of the glasses existence state in processed face image. 6. The method of claim 5, wherein the step of controlling recognition methods further comprisesapplying an individual feature space that is the most efficient in recognizing a person wearing glasses so as to apply to a recognition algorithm when there are glasses. 7. The method of claim 1, wherein the identifying a figure according to a watch list further comprises applying an infrared measurement process and calculating a distance between an input shape and a template shape to identify the figure. 8. The method of claim 1, wherein the identifying a figure according to a watch list further comprises applying a plurality of feature spaces and providing a plurality of input shapes that are generated according to various fragments of a face image to identify the figure. 9. The method of claim 8, wherein the identifying a figure according to a watch list further comprises calculating infrared similarity measurement values for stored input shape images by using corresponding templates from a database, and generating general measurement values by summing the calculated values. 10. The method of claim 9, wherein the calculating infrared similarity measurement values of various images for the input shape further comprises applying linear sums of numbers sequentially acquired by using Support Vector Method and an AdaBoost algorithm. 11. The method of claim 10, wherein when the Support Vector Method is used so as to combine the similarity measurement values,initial vectors of the similarity measurement values are based on generating a separation curve between branches of two kinds acquired from conversion of ordered sequences, and each order sequence includes similarity values that are acquired by different recommendations for one class. 12. A figure recognition system for automatically detecting, tracking, and recognizing a face image by using a watch list, the system comprising: a monitoring scene remote information collection device for transmitting a 2D image taken by a camera to a face detection and tracking device;a face detection and tracking device including a face and face feature detection blocks, for searching for a face image, measuring a detected face feature position coordinate, and tracking the face that is found in a subsequent frame during a video sequence process; a face image quality estimation device including an image feature estimation block and a face feature estimation block, for computing a parameter value displayed on the face image provided by the detection device and transmitting a result to a recognition method control device;a recognition method control device including a face recognition method repository and a method selection block, for selecting a recognition method that is the most efficient for a current face image according to a quality estimate provided by the face image quality estimation device;a biometric feature formation device including a preprocess block and a coding block, for generating a concise face image that is processed by using the selected recognition method;a figure recognition device including a similarity measurement calculation block and a determination selection block, and measuring similarity of various processed face images, summing acquired values, and determining a similarity between a template acquired from a database and an input shape;a database management system for template databases of persons given in a watch list; and a watch list adaptation device for controlling a recognition algorithm based on analysis of elements and structure of the watch list.
연구과제 타임라인
LOADING...
LOADING...
LOADING...
LOADING...
LOADING...
이 특허에 인용된 특허 (2)
Liu, Chengjun; Wechsler, Harry, Feature based classification.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.