WO2020073709A1 - 一种多摄像机多人脸视频接续采集装置及方法 - Google Patents

一种多摄像机多人脸视频接续采集装置及方法 Download PDF

Info

Publication number
WO2020073709A1
WO2020073709A1 PCT/CN2019/096716 CN2019096716W WO2020073709A1 WO 2020073709 A1 WO2020073709 A1 WO 2020073709A1 CN 2019096716 W CN2019096716 W CN 2019096716W WO 2020073709 A1 WO2020073709 A1 WO 2020073709A1
Authority
WO
WIPO (PCT)
Prior art keywords
face
sequence
image
camera
video
Prior art date
Application number
PCT/CN2019/096716
Other languages
English (en)
French (fr)
Inventor
俞杰
俞江峰
朱伟平
石旭刚
Original Assignee
杭州中威电子股份有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 杭州中威电子股份有限公司 filed Critical 杭州中威电子股份有限公司
Priority to US16/838,024 priority Critical patent/US11216645B2/en
Publication of WO2020073709A1 publication Critical patent/WO2020073709A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T3/00—Geometric image transformations in the plane of the image
    • G06T3/40—Scaling of whole images or parts thereof, e.g. expanding or contracting
    • G06T3/4038—Image mosaicing, e.g. composing plane images from plane sub-images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/20—Analysis of motion
    • G06T7/292—Multi-camera tracking
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/40—Scenes; Scene-specific elements in video content
    • G06V20/49—Segmenting video sequences, i.e. computational techniques such as parsing or cutting the sequence, low-level clustering or determining units such as shots or scenes
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/161—Detection; Localisation; Normalisation
    • G06V40/166—Detection; Localisation; Normalisation using acquisition arrangements
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/168—Feature extraction; Face representation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
    • G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
    • G06V40/16—Human faces, e.g. facial parts, sketches or expressions
    • G06V40/172—Classification, e.g. identification
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/30—Subject of image; Context of image processing
    • G06T2207/30196—Human being; Person
    • G06T2207/30201—Face

Definitions

  • the invention relates to video image processing technology and face collection and recognition technology.
  • IP cameras have been widely used, and face AI technology is also very mature, but most face applications are limited to face capture and face comparison.
  • Some technologies started to involve the camera aiming at a certain face to continuously capture video and extract feature points from the video clips of the face to perform time-series waveform processing and further in-depth analysis. However, these technologies all use the camera to aim at a certain face for continuous face video collection.
  • the collection of single time points or non-continuous scenes is usually a face of a certain period of time, and it is easy to deform or the recognition is not high.
  • the existing collection technology is obviously unable to meet the recognition that is increasingly improved with the development of technology Requirements, the application scenarios of the collected images cannot be expanded.
  • the first technical solution to be solved by the present invention is to provide a multi-camera multi-face video continuous acquisition device, which can realize accurate and continuous acquisition of facial images, and can provide an accurate and reliable basis for subsequent analysis.
  • Multi-camera multi-face video continuous acquisition device including,
  • At least one camera for intermittent or continuous collection of continuous or intermittent videos or images with human faces
  • At least one stitching server used for face tracking, face recognition, face cutting, face sorting and face stitching on face videos or images collected by one or several cameras
  • At least one time synchronizer used to calibrate the time of at least one camera and splicing server
  • the above-mentioned devices are interconnected through the network, so as to realize data exchange between each other.
  • one camera can also be used.
  • multiple cameras can be placed at different angles and positions for shooting at different angles, so as to obtain more comprehensive image information.
  • the face video refers to a video that includes the face's active state.
  • the face image refers to an image that includes various decomposition and active states of the face.
  • a series of The human face image may constitute a human face video.
  • the human face image may also contain other objects or other parts of the human body.
  • face recognition refers to the recognition of different faces by the pointer, the purpose of which is to distinguish different faces and thereby identify individual differences.
  • face recognition refers to the element of the face to the face The purpose of identification is to distinguish a person's face from other body parts, so as to judge the number of faces or serve as a reference for other judgment standards.
  • face tracking, face recognition, face cutting, face sorting, and face splicing are performed successively.
  • face tracking, face recognition, face cutting, face sorting, and face splicing are performed sequentially in order.
  • the above operations are directed to multiple faces.
  • such face tracking, face recognition, face cutting, face sorting and face splicing are performed independently.
  • the above operations can be performed relatively independently, for example, face tracking and face recognition can be performed separately, and face cutting and face recognition can be performed separately, provided that face cutting has received the required
  • face tracking and face recognition can be performed separately, and face cutting and face recognition can be performed separately, provided that face cutting has received the required
  • the cut material in some preferred ways, the face
  • the splicing server performs the above steps and finally obtains video and / or images with connection features containing at least one face.
  • the present invention also provides a multi-camera multi-face video continuous acquisition method, which uses the above device and includes the following steps:
  • Image acquisition multiple cameras continuously acquire videos or images with faces in a certain scene
  • the stitching server tracks and recognizes the face of the video or image
  • the stitching server performs face cutting frame by frame to obtain the cut face image
  • Face sorting sort face images according to different timestamps saved in each face image to get face sequences
  • Face comparison match the face images in the face sequence with the face images in the face comparison library, and store the unmatched face sequences as new data in the database;
  • Face splicing the face sequence matched in the face comparison library and the sequence in the library are spliced in time sequence to form a new face sequence.
  • the step (1) also includes the step of decoding the video stream. Specifically, it means extracting the video stream collected by the camera, decoding it, generating each frame of image, and recording the time stamp for each frame of image.
  • the time stamp can be recognized by the time synchronizer.
  • the video stream decoding may include only the step of decoding.
  • the video stream decoding may include the generation of frame images.
  • the video stream decoding may also include the frame
  • the image is marked with a timestamp.
  • the timestamp can be recognized by the time synchronizer.
  • this timestamp will always follow the marked frame image.
  • the face sorting step the faces are sorted according to the marked time stamps. In some preferred methods, face stitching stitches the faces in sequence according to the marked time stamps.
  • the recognition in step (3) specifically refers to performing multi-face detection on each frame of image, individually identifying each detected face, and extracting feature point coordinates for the recognized face .
  • the identification refers to different markings on different faces, so that different faces can be distinguished in subsequent steps.
  • the feature points refer to The difference between each face and other faces.
  • the tracking in step (3) refers to that, after recognizing a certain person's face, when performing face recognition on each subsequent frame of images, it is necessary to recognize whether the next frame of the image contains the face, If it does, it will continue to extract the feature point coordinates, if it does not, it will be identified as a new face, and it will continue to be recognized in other images after it.
  • the unique mark on the face will be brought into the subsequent steps.
  • the tracking can identify whether the recognized human face is included in the other frame images according to the feature points. In some preferred ways,
  • the face cutting in step (4) specifically refers to cutting out the face identified in each frame image from the large image of the video frame, generating a small image of a single face, and copying the frame image Timestamp.
  • the small graphics of a single face there is a one-to-one correspondence between the small graphics of a single face and the large image cut out from the video frame.
  • the face sorting in step (5) specifically refers to sorting the small images of the same person's face in chronological order, which is called the face sequence of the face, and select one of the face sequences As a face comparison chart.
  • the face comparison in step (6) specifically refers to comparing the face comparison image in the face sequence with the face comparison image in the face comparison library to confirm whether it matches, and if it matches ( This matching is equivalent to comparing this face comparison image with each image in the face library.
  • the principle of comparison is to extract several features of the face and model to generate a vector with several features.
  • the face pictures of the comparison library are also modeled to generate vectors in advance, and then the vector set closest to the vector of this face comparison picture is calculated, and sorted according to similarity.
  • a threshold can be set, and the similarity is greater than How many% is considered to match, for example, if the similarity is greater than 75%, it is considered to be a match, the specific value of this similarity can be determined according to need, or can be adjusted according to the actual situation, if there are multiple choices with the closest similarity ,
  • the face sequence and the corresponding face sequence in the face comparison library belong to the same face. If they do not match, the face sequence is considered to be a new face. At this time, the face sequence is added to the face comparison In the library.
  • the face stitching in step (7) specifically means that if the current face sequence matches the face sequence in the face comparison library, the two face sequences are considered to belong to the same face, and the current person is considered
  • the face sequence and the face sequence in the face comparison library are stitched in time sequence to form a new face sequence, and the new face sequence is associated with the face sequence in the face comparison library.
  • the method of the present invention further includes the step of performing time-series waveform analysis on the face sequence stitched multiple times.
  • the time-series waveform analysis refers to whether Continuous waveforms are stitched together to form a complete long-period waveform.
  • the collected face sequence may be discontinuous.
  • the face sequence may be considered to be discontinuous.
  • the method according to the present invention may form multiple discontinuous sequences for the same face. In this case, each small figure of a single face may be regarded as an element in the sequence. There may be multiple repetitive elements in the discontinuous face sequence of the face.
  • these discontinuous face sequences can be stitched into a continuous face sequence for the same face through time-series waveform analysis.
  • the splicing is performed according to the period of the waveform.
  • time-series waveform analysis includes selecting a certain area of each face of each frame, and extracting a certain number of bit values for each color in each pixel in this area. , Forming a number, in some preferred ways, the selected area can be surrounded by multiple feature points, and then the number generated by each pixel of this area is averaged to get a value of this frame, and then The time-stamp of each frame is arranged horizontally to form a time-series waveform.
  • the color of the pixel is RGB color.
  • the pixels of the picture can be represented by three colors of red, green, and blue respectively or in combination.
  • the digital aspect is used. In other words, each color can be represented by 8bi or 16bit.
  • the time-series waveform analysis is to screen and reorganize the already obtained face sequences in order to obtain a more complete face splicing sequence.
  • filtering operations can also be performed on long-period waveforms. The accuracy of the long-period waveform is ensured through the filtering operation, so as to ensure the order and accuracy of the face sequence stitched according to the long-period waveform.
  • the purpose of the present invention is to use a network of several cameras to unconsciously collect several personal faces at the same time and stitch each individual face in time series.
  • a longer-term face sequence can be formed.
  • the face sequence sorted by time series can further extract feature information for various time-series analysis, and The longer the length of the face sequence, the more effective information can be extracted after time series analysis.
  • several cameras are successively installed in the security inspection long channel. When the security personnel pass through these cameras, they will be captured by the camera, and these facial sequence fragments may be stitched together for analysis of emotional stress.
  • several cameras are installed in an open old-age care place. When the old-age staff walks in the open place, several fragments are collected by these cameras, and these facial sequence fragments can be combined to analyze various signs.
  • face sequences can be stitched without limitation.
  • FIG. 1 is a schematic diagram of the overall structure of the present invention.
  • FIG. 3 is a schematic diagram of face tracking recognition of the present invention.
  • FIG. 4 is a schematic diagram of face ordering of the present invention.
  • FIG. 5 is a schematic diagram of face stitching of the present invention.
  • FIG. 6 is a schematic diagram of sequential waveform stitching.
  • FIG. 7 is a schematic diagram of feature point selection for multi-face detection.
  • Embodiment 1 a multi-camera multi-face video continuous acquisition device.
  • the present invention may include multiple cameras for successively capturing videos or images with human faces, for example, camera 1, camera 2, ..., camera n, and these cameras may be the same or different configurations However, all of them can have the function of continuously collecting images or videos. In some preferred ways, these cameras can be connected to the server or the cloud through wired or wireless means to achieve interaction with other devices.
  • the present invention may further include at least one splicing server.
  • at least one splicing server In this embodiment, two splicing servers are included.
  • the splicing server is used for face tracking and tracking of face videos or images collected by the camera. Recognition, face cutting, face sequencing and face stitching;
  • one splicing server can process the video and / or images input from 5-10 cameras at the same time.
  • the processing capacity of the splicing server can be expanded or the number of splicing servers can be increased.
  • the number of splicing servers and the number of cameras are satisfied, and the splicing servers are just capable of processing the video or image collected by the cameras.
  • At least one time synchronizer used to calibrate the time of the camera and the stitching server; in some preferred ways, the time synchronizer can be used to add a time stamp to each frame of the camera and the stitching server, in some preferred In the mode, the time synchronizer can recognize these timestamps. In some preferred modes, the time synchronizer feeds back the recognition result to the stitching server. In some preferred modes, the time synchronizer includes one. In some preferred modes Among them, there are multiple time synchronizers, and multiple time synchronizers use the same format time stamp.
  • the above-mentioned devices are interconnected through the network, so as to realize data exchange between each other.
  • the network may be an internal network with a security code. In some preferred modes, the network may be a public network. In some preferred modes, the network may be a cloud. In some preferred modes, The network can be expanded.
  • Embodiment 2 a multi-camera multi-face video continuous acquisition.
  • the present invention uses the multi-camera and multi-face video continuous acquisition device in Embodiment 1 for splicing, and specifically includes the following steps:
  • Image acquisition multiple cameras continue to acquire video or images with faces in a certain scene, extract the video stream collected by the camera, decode, generate each frame of image, and record each frame of image Time stamp, which can be recognized by the time synchronizer.
  • the timestamp can be generated by a time synchronizer. In some preferred ways, the timestamp can also be an external input and can be recognized by the time synchronizer. In some preferred ways, one time connection For collection, a unified standard time stamp is used. In some preferred methods, the time stamp will always follow the marked frame image once it is marked. In some preferred modes, the timestamp can be replaced with other forms of marks, which can be used to mark the collected and decoded frame images in order, and the marks can be recognized by the time synchronizer.
  • the function of the time synchronizer is to ensure that the time of the images in the camera and the stitching server are consistent.
  • the sequence of the images can be calibrated in the form of marks. The other mark module replaces the time synchronizer.
  • the time synchronizer can be used to calibrate the time of the camera and the splicing server. In some preferred ways, the time synchronizer is separated by a certain time. Calibrate the time of the splicing server.
  • the camera sends the collected video or image with connection characteristics to the splicing server, and the splicing server performs subsequent processing.
  • it may be a continuous picture or process in a certain scene, that is, continuously collect images in a certain process, these images are actually different pictures of a continuous video at each time point, we call this Of the images are connected.
  • the image can be sent in real time, in some preferred ways, the image can be stored in the camera, and in some preferred ways, the image can be stored in the stitching server. In some preferred modes, the stored images can be retrieved. In some preferred modes, the camera can store the images while sending the images. In some preferred modes, the stitching server can store the images before and after stitching.
  • the stitching server tracks and recognizes faces of videos or images. Recognition specifically refers to multi-face detection for each frame of image x.
  • the face detection algorithm is also relatively mature now. There are many, we can use the existing technology, in the existing technology, there may be the situation of face overlap, such as shown in Figure 7, at this time, in fact, the feature points of the face can also be extracted, according to the face The general characteristics of the face, such as the position of the face, which determine the position of the basic elements of the face, can be verified in the subsequent steps.
  • the judgment of the position is verified to be correct, it is adopted and can be used as a learning
  • the feature point can be excluded, and the next time it can be extracted, it can be pre-judgment and verification.
  • it is a coincident feature point, in the process of timing , You can exclude the overlapping feature points or the feature points of the occlusion parts, in addition, if the angle of the face is too large, it may cause features Repeating or missing feature points, because the angle of the face is too large, it needs to be eliminated.
  • the so-called multi-face detection is to detect all the faces in the picture at the same time, rather than just identifying a certain person in a certain picture.
  • Face and uniquely identify each detected face Y of each person, and extract the feature point coordinates of the recognized face; tracking refers to, after identifying a certain person's face, in each subsequent frame
  • face recognition it is necessary to recognize whether the next image x 1 contains the face Y. If it contains, then continue to extract the feature point coordinates. If it does not, it is identified as the new face Y 1 .
  • the selection of feature points can be added or deleted according to actual needs.
  • the feature points include a part where a certain face may be different from other faces, such as the eye, For cheeks, in some preferred ways, feature points may include certain elements that may cause a difference between a face and other faces, such as the size of the nose tip, the height of the bridge of the nose, etc.
  • the feature points may be It is an element that can be easily identified in image recognition.
  • the feature points can be individual elements or a combination of elements.
  • the recognition priority of the feature points can be set, for example, Prioritize certain elements, such as eye size, when these elements are not enough to distinguish a certain face from other faces, and then further identify other elements. And so on.
  • the stitching server performs face cutting frame by frame to obtain the cut face image y; specifically, the face recognized in each frame image x is cut from the large image Y of the video frame Come out, generate a small image y of a single face, and copy the timestamp of the frame image x.
  • the timestamp can be used as the basis for subsequent sorting and stitching.
  • this method in a video collected, there will be multiple frames, Each frame corresponds to a small image y.
  • y 1 , y 2 , y 3 , y 4 , ..., y n are formed for a face Y, where n is a constant and can be determined according to the number of video frames collected or recognized.
  • Face sorting sort face images according to different timestamps saved in each face image y 1 , y 2 , y 3 , y 4 , ..., y n , as shown in Figure 4, press
  • the face sequence is obtained in time sequence.
  • a better quality image y a can be selected as the face comparison image, which can be used for subsequent comparison.
  • Face comparison match the face images in the face sequence y 1 , y 2 , y 3 , y 4 , ..., y n with the face images in the face comparison library, and match Face sequences that are not available are stored as new data; specifically, the face comparison graph y a in the face sequences y 1 , y 2 , y 3 , y 4 , ..., y n is compared with the face Compare the face comparison graph y b of all face sequences in the library (b is just a code, used to distinguish it from a, without special meaning), confirm whether it matches, and if it matches a suitable face sequence z (z contains z 1 , z 2 , z 3 , z 4 , ..., z n , the face comparison graph of z can be z a ), then the face sequence and the face comparison library corresponding face in the library The sequences belong to the same face. If they do not match, the face sequence belongs to the new face.
  • Face stitching the face sequences matched in the face comparison library and the sequences in the library are stitched in chronological order to form a new face sequence.
  • face stitching specifically refers to, If the current face sequence y 1 , y 2 , y 3 , y 4 , ..., y n and the face sequence z 1 , z 2 , z 3 , z 4 , ..., z n in the face comparison library If the match is successful, the two face sequences are considered to belong to the same face, and the current face sequence and the face sequences in the face comparison library are stitched in time sequence to form a new face sequence, and the new face sequence
  • the face sequence is related to the face sequence in the face alignment library.
  • the present invention can also perform time-series waveform analysis on multiple spliced face sequences. As shown in FIG. 6, after a face sequence is spliced several times, the time span of the face sequence will be longer, but there is generally a gap in the middle The time period has not been sampled. When the time-series waveform analysis of some feature points of the spliced face sequence will form several disconnected waveforms, several discrete waveforms can be spliced according to the fluctuation period To form a complete long-period waveform. Long-period waveform means that several waveforms are spliced together. When splicing, the splicing is performed according to the waveform period.
  • each area of each face of each frame (this area can be surrounded by multiple feature points) ) RGB of each pixel (the pixels of the picture are represented by a combination of red, green, and blue, each color can be represented by 8bi or 16bit), each color extracts a specific number of bit values Combine them to form a number, and then average the number generated by each pixel in this area to get a value for this frame, and then arrange it horizontally according to the time stamp of each frame to form a time-series waveform. (Sometimes the actual application will involve filtering, etc.).

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Physics & Mathematics (AREA)
  • Multimedia (AREA)
  • Oral & Maxillofacial Surgery (AREA)
  • Health & Medical Sciences (AREA)
  • Human Computer Interaction (AREA)
  • General Health & Medical Sciences (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Computational Linguistics (AREA)
  • Software Systems (AREA)
  • Computing Systems (AREA)
  • Image Analysis (AREA)
  • Image Processing (AREA)
  • Collating Specific Patterns (AREA)

Abstract

本发明提供一种多摄像机多人脸视频接续采集装置，包括，多台摄像机，用于接续采集带有人脸的视频或图像，至少一台拼接服务器，用于对摄像机采集到的人脸视频或者图像进行人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接；至少一台时间同步器，用于对摄像机和拼接服务器的时间进行校准；上述设备通过网络互联，从而实现相互之间的数据交互。本发明通过对多台摄像机采集的同一个人的人脸图像的时序化拼接可以形成更加长时间段的人脸序列，按时间序列排序的人脸序列后续可以进一步提取特征信息进行各种时序化分析，而人脸序列的时间长度越长，时序化分析后能够提取更多有效信息。

Description

一种多摄像机多人脸视频接续采集装置及方法 技术领域
本发明涉及视频图像处理技术和人脸采集识别技术。
背景技术
IP摄像机已经应用非常广泛，人脸AI技术也已经非常成熟，但是大多数的人脸应用局限在人脸抓拍和人脸比对上。有一些技术开始涉及用摄像机对准某个人脸进行持续拍摄视频并对人脸的视频片段提取特征点进行时间序列的波形处理，并进一步深度分析。但是这些技术都是采用摄像机对准某个人脸进行持续人脸视频采集的。而单时间点或者非连续场景的采集，通常是某个时间段的人脸，很容易发生变形或者识别度不高的情况，现有的采集技术显然是无法满足随着科技发展日益提高的识别要求，也无法扩展所采集到的图像的应用场景。
发明内容
本发明首先要解决的技术方案是提供一种多摄像机多人脸视频接续采集装置，能够实现人脸图像的准确和接续采集，并可为后续的分析提供准确可靠的基础。
为此，本发明采用以下技术方案：
一种多摄像机多人脸视频接续采集装置，包括，
至少一台摄像机，用于间断或持续采集带有人脸的连续或间隔一定时间的视频或者图像，
至少一台拼接服务器，用于对其中一台或某几台摄像机采集到的人脸视频或者图像进行人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接，
至少一台时间同步器，用于为至少一台摄像机和拼接服务器的时间进行校准；
上述设备通过网络互联，从而实现相互之间的数据交互。
在一些优选的方式中，摄像机也可以采用一台，在一些优选的方式中，多台 摄像机可以采用不同角度和位置进行摆放以便进行不同角度的拍摄，从而获取更加全面的图像信息。
在一些优选的方式中，人脸视频是指包含人脸活动状态的视频，在一些优选的方式中，人脸图像是指包含人脸各个分解活动状态的图像，在一些优选的方式中，一连串的人脸图像可以构成人脸视频，在一些优选的方式中，人脸图像也可以包含其他的物体或者人体的其他部位。
在一些优选的方式中，人脸识别是指针对不同人脸的识别，其目的是区别不同的人脸，从而识别个体差异，在一些优选的方式中，人脸识别是指针对人脸这个要素的识别，其目的是将人的面部和其他肢体部位区别开来，从而判断人脸数量或者作为其他判断标准的参考。
在一些优选的方式中，这种人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接是接续进行的。在一些优选的方式中，人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接按照顺序依次进行，在一些优选的方式中，上述的这些操作均针对多个人脸。
在一些优选的方式中，这种人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接是分别独立进行的。在一些优选的方式中，上述这些操作可以相对独立进行，例如，人脸跟踪和人脸识别可以分别进行，人脸切割和人脸识别可以分别进行，前提是，人脸切割已经收到了所需要切割的素材，在一些优选的方式中，人脸
在一些优选的方式中，拼接服务器执行上述步骤并最终获得包含至少一个人脸的具有接续特点的视频和/或图像。
同时，本发明还提供一种多摄像机多人脸视频接续采集方法，该方法采用上述的装置，并包括以下步骤：
(1)图像采集：多台摄像机对某一场景下带有人脸的视频或图像进行接续采集；
(2)图像发送：摄像机将采集到的具有接续特点的视频或图像发送至拼接服务器；
(3)跟踪识别：拼接服务器对视频或图像进行人脸的跟踪和识别；
(4)人脸切割：拼接服务器逐帧进行人脸切割，得到切割后的人脸图像；
(5)人脸排序：根据每个人脸图像所保存的不同的时间戳进行人脸图像的排序，得到人脸序列；
(6)人脸比对：将人脸序列中的人脸图像与人脸比对库中的人脸图像进行匹配，将匹配不到的人脸序列作为新的数据入库；
(7)人脸拼接：将人脸比对库中匹配到的人脸序列与库中序列按照时间顺序进行拼接，形成新的人脸序列。
进一步地，在步骤(1)中还包括视频流解码的步骤，具体是指，提取摄像机所采集到的视频流，进行解码，生成每一帧图像，并对每一帧图像记录时间戳，该时间戳能够被时间同步器识别。在一些优选的方式中，视频流解码可以是只包括解码的步骤，在一些优选的方式中，视屏流解码可以包括帧图像的生成，在一些优选的方式中，视频流解码还可以包括对帧图像进行时间戳的标记，在一些优选的方式中，所述的时间戳能够被时间同步器识别，在一些优选的方式中，这个时间戳会一直跟随所标记的帧图像，在一些优选的方式中，在人脸排序步骤中，根据所标记的时间戳来排序人脸，在一些优选的方式中，人脸拼接根据所标记的时间戳来依次拼接人脸。
进一步地，步骤(3)中的识别具体是指，对每一帧图像进行多人脸检测，并对检测到的每一个人脸分别进行唯一标识，对于识别到的人脸进行特征点坐标提取。在一些优选的方式中，所述标识是指对不同的人脸进行不同标记，使得不同的人脸在后续的步骤中可以被区分开来，在一些优选的方式中，所述特征点是指每个人脸和其他人脸的区别点。
进一步地，步骤(3)中的跟踪是指，在识别到某一个人脸之后，在对其后的每一帧图像进行人脸识别时，均需要识别下一帧图像是否包含该人脸，如果包含，则继续提取特征点坐标，如果不包含，则标识为新的人脸，在其后的其他图像中继续进行识别。在一些优选的方式中，对人脸的唯一标记会被带入到其后的步骤中。在一些优选的方式中，所述的跟踪可以根据特征点来识别其他帧图像中是否包含已经识别出的人脸。在一些优选的方式中，
进一步地，步骤(4)的人脸切割具体是指，将每一帧图像中识别到的人脸从视频帧的大图像中切割出来，生成单个人脸的小图像，并拷贝该帧图像的时间戳。在一些优选的方式中，单个人脸的小图形和视频帧中切割出来的大图像是一 一对应的关系。
进一步地，步骤(5)的人脸排序具体是指，将同一个人脸的切割出来的小图像，按照时间顺序进行排序，称为该人脸的人脸序列，在人脸序列中选取一张作为人脸比对图。
进一步地，步骤(6)的人脸比对具体是指，将人脸序列中的人脸对比图与人脸比对库中的人脸比对图进行比对，确认是否匹配，如果匹配(这个匹配相当于用这张人脸比对图与人脸库中的每一张图进行比对，比对的原理是提取人脸的若干个特征建模生成具有若干个特征的向量，人脸比对库的人脸图片也都事先建模生成向量，然后计算跟这张人脸比对图的向量距离最近的向量集合，按照相似度进行排序，一般可以设置一个阀值，认为相似度大于多少％认为是匹配的，例如，相似度大于75％的认为是匹配，这个相似度的具体数值可以根据需要确定，或者也可以根据实际情况进行调整，如果有多个选择相似度最接近的)，则认为人脸序列与人脸比对库中对应的人脸序列属于同一个人脸，如果不匹配，则认为人脸序列属于新的人脸，此时将人脸序列加入人脸比对库中。
进一步地，步骤(7)的人脸拼接具体是指，如果当前人脸序列和人脸比对库中的人脸序列匹配成功，则认为这两个人脸序列属于同一个人脸，则将当前人脸序列和人脸比对库中的人脸序列按照时间顺序进行拼接，形成新的人脸序列，并将该新的人脸序列关联到人脸比对库中的人脸序列。
在一些优选的方式中，本发明的方法还包括对多次拼接的人脸序列进行时序化波形分析的步骤，在一些优选的方式中，所述的时序化波形分析是指根据波动周期对不连续的波形进行拼接，形成完整的长周期波形。在一些优选的方式中，所采集的人脸序列可能是不连续的，例如，所采集的人脸序列拼接后无法形成完整的人脸动作，此时可以认为人脸序列是不连续的，在一些优选的方式中，根据本发明的方法可能会形成多个针对同一人脸的不连续的序列，此时每一个单个人脸的小图形可以认为是序列中的一个元素，在多个针对同一人脸的不连续的人脸序列中，可能会存在多个重复的元素，此时可以通过时序化波形分析，将这些不连续的人脸序列拼接为针对同一人脸的连续的人脸序列，在一些优选的方式中，在将几段波形拼接起来的时候照波形周期进行拼接。
在一些优选的方式中，时序化波形分析包括选定每一帧每一个人脸的某个区 域，将这个区域中的每一个像素中的每一种颜色提取特定几个bit位的值组合起来，形成的一个数字，在一些优选的方式中，所选定的这个区域可以用多个特征点框围，然后对这个区域的每一个像素生成的这个数字进行平均得到这帧的一个数值，然后按每一帧的时间戳进行横向排列形成一个时序化波形。在一些优选的方式中，像素的颜色是RGB颜色，在一些优选的方式中，图片的像素可以用红、绿、蓝三色分别表示或者组合表示，在一些优选的方式中，就数字方面来说，每一种颜色可以分别由8bi或者16bit来表示。
在一些优选的方式中，可以认为时序化波形分析是对已经获得的人脸序列的筛选和重组，以期获得更为完整的人脸拼接序列。在一些优选的方式中，对长周期波形还可以进行滤波操作。通过滤波操作来确保长周期波形的准确性，从而确保根据长周期波形进行拼接的人脸序列的有序性和准确性。
本发明的有益效果是：本发明的目的是通过联网的若干台摄像机无感知的对若干个人脸同时进行采集并对每一个人脸按时间序列进行拼接的技术。通过对多台摄像机采集的同一个人的人脸图像的时序化拼接可以形成更加长时间段的人脸序列，按时间序列排序的人脸序列后续可以进一步提取特征信息进行各种时序化分析，而人脸序列的时间长度越长，时序化分析后能够提取更多有效信息。比如在安检长通道中接续安装若干台摄像机，当被安检人员经过这些摄像机时都会被摄像机采集到片段，将这些人脸序列片段拼接后可以进行情绪压力等的分析。比如在开放的养老场所安装若干台摄像机，当养老人员在开放场所走动时，被这些摄像机采集到若干个片段，将这些人脸序列片段拼接后可以进行各种体征指标的分析。理论上人脸序列可以无限制的拼接。
附图说明
图1是本发明的整体结构示意图。
图2是本发明的整体流程图。
图3本发明的人脸跟踪识别的示意图。
图4是本发明的人脸排序的示意图。
图5是本发明的人脸拼接的示意图。
图6是时序化波形拼接示意图。
图7是多人脸检测的特征点选取示意图。
图8是特征点的示意图。
具体实施方式
以下结合附图对本发明的技术方案做进一步详细说明，应当指出的是，实施例只是对本发明的具体阐述，不应视为对本发明的限定。具体实施方式中可能会包含多个实施例，每个实施例均可以独立实现或者组合实现本发明的全部或部分的技术方案，实施例可以是完整的技术方案，也可以是本发明的一个环节，不应将实施例视为本发明的全部，应当理解的是，对于本领域普通技术人员来说，在不付出创造性劳动的前提下，还可以根据这些实施方式获得其他的实施方式。
在具体实施方式中所引用的附图，仅仅是为了更好地说明本发明的实施例和展现本发明的其中一种或多种实现方式，应当理解的是，对于本领域普通技术人员来说，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
实施例1，一种多摄像机多人脸视频接续采集装置。
参照附图1。
如图1所示，本发明可以包括，用于接续采集带有人脸的视频或图像的多台摄像机，例如，摄像机1、摄像机2、……、摄像机n，这些摄像机可以是相同或者不同的配置，但是都可以具备持续采集图像或者视频的功能，在一些优选的方式中，这些摄像机可以通过有线或者无线的方式连接至服务器或者云端，从而实现和其他设备的交互。
如图1所示，本发明还可以包括至少一台拼接服务器，在本实施例中，包括两台拼接服务器，拼接服务器用于对摄像机采集到的人脸视频或者图像进行人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接；
在一些优选的方式中，一台拼接服务器可以同时处理5台-10台摄像机输入的视频和/或图像，在一些优选的方式中，可以扩大拼接服务器的处理量，或者增加拼接服务器的数量，在一些优选的方式中，拼接服务器的数量和摄像机的数量满足，拼接服务器恰好能够完成摄像机所采集的视频或者图像的处理。
至少一台时间同步器，用于对摄像机和拼接服务器的时间进行校准；在一些优选的方式中，时间同步器可以用于对摄像机和拼接服务器中的每一帧图像添加 时间戳，在一些优选的方式中，时间同步器可以识别这些时间戳，在一些优选的方式中，时间同步器将识别结果向拼接服务器反馈，在一些优选的方式中，时间同步器包括一台，在一些优选的方式中，时间同步器包括多台，多台时间同步器采用同一个格式的时间戳。
上述设备通过网络互联，从而实现相互之间的数据交互。
在一些优选的方式中，网络可以是带有安全码的内部网络，在一些优选的方式中，网络可以是公共网络，在一些优选的方式中，网络可以是云端，在一些优选的方式中，网络可以扩展。
实施例2，一种多摄像机多人脸视频接续采集。
参照附图2。
如图2所示为本发明的一种实现方法，在这种实现方法中，本发明采用实施例1中的多摄像机多人脸视频接续采集装置进行拼接，具体包括如下步骤：
(1)图像采集：多台摄像机对某一场景下带有人脸的视频或图像进行接续采集，提取摄像机所采集到的视频流，进行解码，生成每一帧图像，并对每一帧图像记录时间戳，该时间戳能够被时间同步器识别。
在一些优选的方式中，时间戳可以由时间同步器来生成，在一些优选的方式中，时间戳也可以是外部输入，并能够被时间同步器所识别，在一些优选的方式中，一次接续采集，使用统一标准的时间戳，在一些优选的方式中，时间戳一旦标记后会始终跟随被标记的帧图像。在一些优选的方式中，时间戳可以替换为其他形式的标记，该标记可以按照先后顺序对采集和解码后的帧图像进行标记，该标记能够被时间同步器识别。
在一些优选的方式中，时间同步器的作用是确保摄像机和拼接服务器中的图像的时间一致，在一些优选的方式中，可以采用标记的形式来标定图像的先后顺序，此时可以以相同功能的其他标记模块代替时间同步器，在一些优选的方式中，时间同步器可以用于校准摄像机和拼接服务器的时间，在一些优选的方式中，时间同步器间隔一定时间，对每台摄像机和每台拼接服务器的时间进行校准。
(2)图像发送：摄像机将采集到的具有接续特点的视频或图像发送至拼接服务器，由拼接服务器进行后续处理。在本实施例中可以是某一个场景中连续不断的画面或过程，也即在某一个过程中不断采集图像，这些图像实际上是一个连 续的视频在每个时间点的不同画面，我们称这样的图像为具有接续特点。
在一些优选的方式中，图像可以实时被发送，在一些优选的方式中，图像可以被存储在摄像机中，在一些优选的方式中，图像可以被存储在拼接服务器中。在一些优选的方式中，存储的图像可以被调取，在一些优选的方式中，摄像机可以在发送图像的同时，存储图像，在一些优选的方式中，拼接服务器可以分别存储拼接前后的图像。
(3)跟踪识别：拼接服务器对视频或图像进行人脸的跟踪和识别，其中，识别具体是指，对每一帧图像x进行多人脸检测，人脸检测的算法现在也比较成熟，方法有比较多，我们可以采用现有技术，在现有技术中，可能会存在人脸重合的情况，例如图7所示，这个时候其实人脸的特征点也是可以提取到的，可以根据人脸的大致特征，例如人脸无关的位置，对人脸基本要素的位置进行判断，并可以在后续步骤中进行验证，如果验证该位置的判断是正确的，则进行采纳，并可以作为一种学习过程，如果在后续步骤中验证是错误的，那么可以将该特征点排除掉，在下一次提取的时候，可以再次进行预判和验证，此外，如果是重合的特征点，在时序化的过程中，可以将重合的特征点或者遮挡部位的特征点排除，此外，如果人脸角度太大的话，也可能会造成特征点重复或者特征点缺失，因为人脸角度太大，也需要排除掉的，所谓的多人脸检测就是对画面中所有的人脸都同时进行检测，而不是仅仅识别某一个画面中的某一个人脸，并对检测到的每一个人脸Y分别进行唯一标识，对于识别到的人脸进行特征点坐标提取；跟踪是指，在识别到某一个人脸之后，在对其后的每一帧图像x进行人脸识别时，均需要识别下一帧图像x 1是否包含该人脸Y，如果包含，则继续提取特征点坐标，如果不包含，则标识为新的人脸Y 1，在其后的其他图像中继续针对Y 1进行识别，也就是说，每当识别出一个新的人脸Y n后，在其后的其他识别都必须对Y n进行后续的继续识别，如图8所示，在ISO/IEC 14496-2(MPEG-4)的规范里定义了一些人脸的特征点，比如鼻尖眼眶一圈的几个点等等，但是实际上不同的业务会对特征点进行扩充，目前我们用到的点在MPEG-4中都有定义。
在一些优选的方式中，特征点的选取，可以根据实际需要增加或删减，在一些优选的方式中，特征点包括某个人脸可能与其他人脸发生差异的某个部位，例如眼部，颊部，在一些优选的方式中，特征点可以包括某个人脸可能与其他人脸 发生差异的某个要素，例如鼻尖的大小，鼻梁的高度等，在一些优选的方式中，特征点的可以是容易在图像识别中被鉴别出来的要素，在一些优选的方式中，特征点可以是单独的要素或者是要素的组合，在一些优选的方式中，可以设置特征点的识别优先级，例如，优先识别某些要素，例如，眼睛大小，当这些要素不足以让某个人脸和其他人脸区别开来时，再进一步识别其他的要素。以此类推。
(4)人脸切割：拼接服务器逐帧进行人脸切割，得到切割后的人脸图像y；具体是指，将每一帧图像x中识别到的人脸从视频帧的大图像Y中切割出来，生成单个人脸的小图像y，并拷贝该帧图像x的时间戳，时间戳可以作为后续排序以及拼接的依据，根据这样的方式，采集到的一段视频中，会有多帧画面，每一帧画面都会对应一个小图像y，当一段视频切割完成后，就会形成针对一个人脸Y的多个小图像y 1、y 2、y 3、y 4、……、y n，其中n为常数，可以根据采集或者识别到的视频帧数确定。
(5)人脸排序：根据每个人脸图像y 1、y 2、y 3、y 4、……、y n所保存的不同的时间戳进行人脸图像的排序，如图4所示，按时间顺序得到人脸序列，在人脸序列中，可以选择一张质量较好的图像y a作为作为人脸比对图，可以用于后续比对使用，y a来源于y 1、y 2、y 3、y 4、……、y n，也即a＝1、2、3、4、……、n，质量较好是指人脸的图像的清晰度高且人脸的转向角度比较小；
(6)人脸比对：将人脸序列y 1、y 2、y 3、y 4、……、y n中的人脸图像与人脸比对库中的人脸图像进行匹配，将匹配不到的人脸序列作为新的数据入库；具体是指，将人脸序列y 1、y 2、y 3、y 4、……、y n中的人脸对比图y a与人脸比对库中所有人脸序列的人脸比对图y b进行比对(b只是一个代号，用于与a区别，无特殊含义)，确认是否匹配，如果匹配到某个合适的人脸序列z(z包含z 1、z 2、z 3、z 4、……、z n，z的人脸比对图可以是z a)，则认为人脸序列与人脸比对库中对应的人脸序列属于同一个人脸，如果不匹配，则认为人脸序列属于新的人脸，此时将人脸序列加入人脸比对库中。
(7)人脸拼接：将人脸比对库中匹配到的人脸序列与库中序列按照时间顺序进行拼接，形成新的人脸序列，如图5所示，人脸拼接具体是指，如果当前人脸序列y 1、y 2、y 3、y 4、……、y n和人脸比对库中的人脸序列z 1、z 2、z 3、z 4、……、z n匹配成功，则认为这两个人脸序列属于同一个人脸，则将当前人脸序列和人脸 比对库中的人脸序列按照时间顺序进行拼接，形成新的人脸序列，并将该新的人脸序列关联到人脸比对库中的人脸序列。
此外，本发明还可以对多次拼接的人脸序列进行时序化波形分析，如图6所示，人脸序列经过若干次拼接后，人脸序列的时间跨度会比较长，但是一般中间有空白时间段没有被采样到，对拼接后的人脸序列的某些特征点位进行时序化波形分析时，会形成若干个不连接的波形，可以根据波动的周期对若干段不连续的波形进行拼接，形成一个完整的长周期波形。长周期波形的意思是将几段波形拼接起来，拼接的时候要按照波形周期进行拼接，具体来说是把每一帧每一个人脸的某个区域(这个区域可以用多个特征点框围)的每一个像素的RGB(图片的像素是用红、绿、蓝三色组合表示的，每一种颜色可以由8bi或者16bit来表示)中的每一种颜色提取特定几个bit位的值组合起来，形成的一个数字，然后对这个区域的每一个像素生成的这个数字进行平均得到这帧的一个数值，然后按每一帧的时间戳进行横向排列形成一个时序化波形。(实际的应用时有时还会涉及滤波等)。
以上所述，仅为发明的具体实施方式，但发明的保护范围并不局限于此，任何不经过创造性劳动想到的变化或替换，都应涵盖在发明的保护范围之内。因此，发明的保护范围应该以权利要求书所限定的保护范围为准。

Claims (15)

  1. 一种多摄像机多人脸视频接续采集装置，其特征是，包括，
    至少一台摄像机，用于间断或持续采集带有人脸的连续或间隔一定时间的视频或者图像，
    至少一台拼接服务器，用于对其中一台或某几台摄像机采集到的人脸视频或者图像进行人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接，
    至少一台时间同步器，用于为至少一台摄像机和拼接服务器的时间进行校准；
    上述设备通过网络互联，从而实现相互之间的数据交互。
  2. 根据权利要求1所述的一种多摄像机多人脸视频接续采集装置，其特征是，这种人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接是接续进行的。
  3. 根据权利要求1所述的一种多摄像机多人脸视频接续采集装置，其特征是，这种人脸跟踪、人脸识别、人脸切割、人脸排序和人脸拼接是分别独立进行的。
  4. 根据权利要求3所述的一种多摄像机多人脸视频接续采集装置，其特征是，拼接服务器执行上述步骤并最终获得包含至少一个人脸的具有接续特点的视频和/或图像。
  5. 一种多摄像机多人脸视频接续采集方法，其特征是，该方法采用权利要求1所述的装置，并包括以下步骤：
    (1)图像采集；
    (2)图像发送：将具有接续特点的视频或图像发送至拼接服务器；
    (3)跟踪识别：对视频或图像进行人脸的跟踪和识别；
    (4)人脸切割：逐帧进行人脸切割，得到切割后的人脸图像；
    (5)人脸排序：根据每个人脸图像所保存的不同的时间戳进行人脸图像的排序，得到人脸序列；
    (6)人脸比对：将人脸序列中的人脸图像与人脸比对库中的人脸图像进行匹配，将匹配不到的人脸序列作为新的数据入库；
    (7)人脸拼接：将人脸比对库中匹配到的人脸序列与库中序列按照时间顺序进行拼接，形成新的人脸序列。
  6. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征 是，所述图像采集和所述图像发送由至少一台摄像机来执行。
  7. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征是，所述跟踪识别、人脸切割、人脸排序、人脸比对和人脸拼接由拼接服务器来执行。
  8. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征是，在步骤(1)中还包括视频流解码的步骤，具体是指，提取摄像机所采集到的视频流，进行解码，生成每一帧图像，并对每一帧图像记录时间戳，该时间戳能够被时间同步器识别。
  9. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征是，步骤(3)中的识别具体是指，对每一帧图像进行多人脸检测，并对检测到的每一个人脸分别进行唯一标识，对于识别到的人脸进行特征点坐标提取。
  10. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征是，步骤(3)中的跟踪是指，在识别到某一个人脸之后，在对其后的每一帧图像进行人脸识别时，均需要识别下一帧图像是否包含该人脸，如果包含，则继续提取特征点坐标，如果不包含，则标识为新的人脸，在其后的其他图像中继续进行识别。
  11. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征是，步骤(4)的人脸切割具体是指，将每一帧图像中识别到的人脸从视频帧的图像中切割出来，生成单个人脸的图像，并拷贝该帧图像的时间戳。
  12. 根据权利要求9所述的一种多摄像机多人脸视频接续采集方法，其特征是，步骤(5)的人脸排序具体是指，将同一个人脸的切割出来的图像，按照时间顺序进行排序，称为该人脸的人脸序列，在人脸序列中选取一张作为人脸比对图。
  13. 根据权利要求10所述的一种多摄像机多人脸视频接续采集方法，其特征是，步骤(6)的人脸比对具体是指，将人脸序列中的人脸对比图与人脸比对库中的人脸比对图进行比对，确认是否匹配，如果匹配，则认为人脸序列与人脸比对库中对应的人脸序列属于同一个人脸，如果不匹配，则认为人脸序列属于新的人脸，此时将人脸序列加入人脸比对库中。
  14. 根据权利要求12所述的一种多摄像机多人脸视频接续采集方法，其特 征是，步骤(7)的人脸拼接具体是指，如果当前人脸序列和人脸比对库中的人脸序列匹配成功，则认为这两个人脸序列属于同一个人脸，则将当前人脸序列和人脸比对库中的人脸序列按照时间顺序进行拼接，形成新的人脸序列，并将该新的人脸序列关联到人脸比对库中的人脸序列。
  15. 根据权利要求5所述的一种多摄像机多人脸视频接续采集方法，其特征是，还包括对多次拼接的人脸序列进行时序化波形分析的步骤，在时序化波形分析中根据波动周期对不连续的波形进行拼接，形成完整的长周期波形。
PCT/CN2019/096716 2018-10-08 2019-07-19 一种多摄像机多人脸视频接续采集装置及方法 WO2020073709A1 (zh)

Priority Applications (1)

Application Number Priority Date Filing Date Title
US16/838,024 US11216645B2 (en) 2018-10-08 2020-04-01 Multi-camera multi-face video splicing acquisition device and method thereof

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN201811166538.4A CN110008797B (zh) 2018-10-08 2018-10-08 一种多摄像机多人脸视频接续采集方法
CN201811166538.4 2018-10-08

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US16/838,024 Continuation US11216645B2 (en) 2018-10-08 2020-04-01 Multi-camera multi-face video splicing acquisition device and method thereof

Publications (1)

Publication Number Publication Date
WO2020073709A1 true WO2020073709A1 (zh) 2020-04-16

Family

ID=67164971

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/096716 WO2020073709A1 (zh) 2018-10-08 2019-07-19 一种多摄像机多人脸视频接续采集装置及方法

Country Status (3)

Country Link
US (1) US11216645B2 (zh)
CN (1) CN110008797B (zh)
WO (1) WO2020073709A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112529778A (zh) * 2020-11-24 2021-03-19 展讯通信（上海）有限公司 多摄像头设备的图像拼接方法及装置、存储介质、终端
CN114093004A (zh) * 2021-11-25 2022-02-25 成都智元汇信息技术股份有限公司 一种基于多摄像头的人脸融合比对方法及装置

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110008797B (zh) * 2018-10-08 2021-12-14 杭州中威电子股份有限公司 一种多摄像机多人脸视频接续采集方法
CN110012256A (zh) * 2018-10-08 2019-07-12 杭州中威电子股份有限公司 一种融合视频通信与体征分析的系统
US11010599B2 (en) 2019-05-01 2021-05-18 EMC IP Holding Company LLC Facial recognition for multi-stream video using high probability group and facial network of related persons
US11334746B2 (en) * 2019-05-01 2022-05-17 EMC IP Holding Company LLC Facial recognition for multi-stream video using high probability group
CN112351327A (zh) * 2019-08-06 2021-02-09 北京字节跳动网络技术有限公司 人脸图像的处理方法、装置、终端及存储介质
US20220075997A1 (en) * 2020-09-07 2022-03-10 Corsight .Ai Face features matching based tracker
CN114140542B (zh) * 2021-12-09 2022-11-22 成都智元汇信息技术股份有限公司 基于色彩补偿的切图方法、电子设备及介质
CN114401378A (zh) * 2022-03-25 2022-04-26 北京壹体科技有限公司 一种用于田径项目的多段视频自动拼接方法、系统和介质
CN114863364B (zh) * 2022-05-20 2023-03-07 碧桂园生活服务集团股份有限公司 一种基于智能视频监控的安防检测方法及系统
CN115076536B (zh) * 2022-06-08 2023-12-19 深圳沃尔克智慧科技有限公司 一种基于人脸识别技术的身份验证装置

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101854516A (zh) * 2009-04-02 2010-10-06 北京中星微电子有限公司 视频监控系统、视频监控服务器及视频监控方法
US8698874B2 (en) * 2011-06-10 2014-04-15 Microsoft Corporation Techniques for multiple video source stitching in a conference room
CN104850607A (zh) * 2015-05-07 2015-08-19 中国人民解放军国防科学技术大学 一种多节点视频重组方法和系统
CN108563651A (zh) * 2017-12-19 2018-09-21 深圳云天励飞技术有限公司 一种多视频的目标搜索方法、装置及设备
CN110008797A (zh) * 2018-10-08 2019-07-12 杭州中威电子股份有限公司 一种多摄像机多人脸视频接续采集装置及方法

Family Cites Families (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US9094675B2 (en) * 2008-02-29 2015-07-28 Disney Enterprises Inc. Processing image data from multiple cameras for motion pictures
CN103237192B (zh) * 2012-08-20 2016-03-30 苏州大学 一种基于多摄像头数据融合的智能视频监控系统
US10169659B1 (en) * 2015-09-24 2019-01-01 Amazon Technologies, Inc. Video summarization using selected characteristics
CN105893965B (zh) * 2016-03-31 2019-03-29 中国科学院自动化研究所 一种可用于非特定人的双目视觉图像合成装置与方法
CN105996993A (zh) * 2016-04-29 2016-10-12 南京理工大学 一种智能视频生命体征监测系统及方法
CN107786803A (zh) * 2016-08-29 2018-03-09 中兴通讯股份有限公司 一种图像生成方法、装置及终端设备
CN108615020A (zh) * 2018-04-28 2018-10-02 东莞市华睿电子科技有限公司 一种视频监控区域内的流动人数统计方法

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN101854516A (zh) * 2009-04-02 2010-10-06 北京中星微电子有限公司 视频监控系统、视频监控服务器及视频监控方法
US8698874B2 (en) * 2011-06-10 2014-04-15 Microsoft Corporation Techniques for multiple video source stitching in a conference room
CN104850607A (zh) * 2015-05-07 2015-08-19 中国人民解放军国防科学技术大学 一种多节点视频重组方法和系统
CN108563651A (zh) * 2017-12-19 2018-09-21 深圳云天励飞技术有限公司 一种多视频的目标搜索方法、装置及设备
CN110008797A (zh) * 2018-10-08 2019-07-12 杭州中威电子股份有限公司 一种多摄像机多人脸视频接续采集装置及方法

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112529778A (zh) * 2020-11-24 2021-03-19 展讯通信（上海）有限公司 多摄像头设备的图像拼接方法及装置、存储介质、终端
CN114093004A (zh) * 2021-11-25 2022-02-25 成都智元汇信息技术股份有限公司 一种基于多摄像头的人脸融合比对方法及装置

Also Published As

Publication number Publication date
US20200234036A1 (en) 2020-07-23
CN110008797B (zh) 2021-12-14
CN110008797A (zh) 2019-07-12
US11216645B2 (en) 2022-01-04

Similar Documents

Publication Publication Date Title
WO2020073709A1 (zh) 一种多摄像机多人脸视频接续采集装置及方法
US7630561B2 (en) Image processing
US7636453B2 (en) Object detection
US8866931B2 (en) Apparatus and method for image recognition of facial areas in photographic images from a digital camera
CN110298231B (zh) 一种用于篮球比赛视频进球判定的方法及系统
CN103353933B (zh) 图像识别设备及其控制方法
CN110264493B (zh) 一种针对运动状态下的多目标对象追踪方法和装置
CN109446981A (zh) 一种脸部活体检测、身份认证方法及装置
CN108830208A (zh) 视频处理方法和装置、电子设备、计算机可读存储介质
JP3617052B2 (ja) カメラ映像提供システム及びカメラ映像の時刻補正方法
CN104361327A (zh) 一种行人检测方法和系统
CN105279480A (zh) 视频分析的方法
TWI382360B (zh) 物件偵測方法及其裝置
US11315360B2 (en) Live facial recognition system and method
CN109961031A (zh) 人脸融合识别认定方法、目标人员信息显示方法、预警布控方法及系统
GB2414616A (en) Comparing test image with a set of reference images
CN108616718A (zh) 监控显示方法、装置及系统
CN101472063A (zh) 一种手机相机捕捉笑脸的方法
CN113065568A (zh) 目标检测、属性识别与跟踪方法及系统
JP2004219277A (ja) 人体検知方法およびシステム、プログラム、記録媒体
TW201328358A (zh) 攝影畫面之物件串接修正方法及其多攝影機監控系統
JPH1115979A (ja) 顔検出及び顔追跡方法並びにその装置
TWI733616B (zh) 人體姿勢辨識系統、人體姿勢辨識方法以及非暫態電腦可讀取儲存媒體
CN104104860B (zh) 对象图像检测设备及其控制方法以及控制程序、记录介质
CN114863337A (zh) 一种新型屏幕防拍照识别方法

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19871647

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19871647

Country of ref document: EP

Kind code of ref document: A1

122 Ep: pct application non-entry in european phase

Ref document number: 19871647

Country of ref document: EP

Kind code of ref document: A1