CN108229442A - Face fast and stable detection method in image sequence based on MS-KCF - Google Patents
Face fast and stable detection method in image sequence based on MS-KCF Download PDFInfo
- Publication number
- CN108229442A CN108229442A CN201810124952.2A CN201810124952A CN108229442A CN 108229442 A CN108229442 A CN 108229442A CN 201810124952 A CN201810124952 A CN 201810124952A CN 108229442 A CN108229442 A CN 108229442A
- Authority
- CN
- China
- Prior art keywords
- detection
- kcf
- face
- image sequence
- target
- Prior art date
- Legal status (The legal status 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 status listed.)
- Granted
Links
Classifications
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/20—Analysis of motion
- G06T7/207—Analysis of motion for motion estimation over a hierarchy of resolutions
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20081—Training; Learning
-
- 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
- G06T2207/20—Special algorithmic details
- G06T2207/20084—Artificial neural networks [ANN]
-
- 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
Landscapes
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- General Health & Medical Sciences (AREA)
- Oral & Maxillofacial Surgery (AREA)
- Human Computer Interaction (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The present invention proposes a kind of detection method of face fast and stable in image sequence based on MS KCF.For angle change in image sequence it is larger, block more serious Face datection problem, the present invention proposes the quick accurate target detection model M obileNet SSD of fusion（MS）With quick trace model core correlation filtering（Kernelized Correlation Filters,KCF）A kind of new automatic detection-tracking-detection（Detection Tracking Detection, DTD）Pattern, i.e. MS KCF Face datections model.This method includes the following steps：Step 1, MS detection networks are built；Step 2, target is detected using MS networks；Step 3, trace model is updated, for predicting the position of next frame human face target；Step 4, after tracking number frame, then MS detection networks are updated, detects positioning again to human face target；Step 5, analysis is compared to experimental result.Experiment show MS KCF models both ensure that it is larger to angle change in image sequence, block the stability of more serious Face datection, while also substantially increase detection speed.
Description
Technical field
The invention belongs to the target detection technique field of machine vision, more particularly to a kind of image sequence based on MS-KCF
Middle face fast and stable detection method.
Background technology
With the continuous development of computer technology, computer performance is continuously improved, and human face detection tech is regarded as computer
An important branch in feel field also achieves huge breakthrough, and nowadays, Face datection is in access control system, intelligent monitoring, intelligence
Energy camera etc. has a wide range of applications.Face datection is also a kind of challenging technology, for angle in image sequence
Change greatly, block more serious face how the detection of real-time stabilization, it has also become urgent problem to be solved in application.Mesh
Before, it can not meet demand, therefore profound convolutional neural networks using the conventional method of shallow-layer feature
（Convolutional Neural Network, CNN）It is the emphasis and hot spot of nowadays detection technique research.
Traditional method for detecting human face is numerous, but all has the characteristics that：First, need artificial selection feature, process
Complexity, the quality of target detection effect depend on the priori of researcher；Second, in a manner that window area traverses image
Detect target, there is many redundancy windows in detection process, time complexity is high, and to angle change in image sequence compared with
Greatly, more serious Face datection less effective is blocked.
In recent years, CNN obtains quantum jump in object detection field, becomes nowadays state-of-the-art target detection side
Method.Significant breakthroughs of the CNN in target detection is the R-CNN that Ross Girshick et al. were proposed in 2014（Region-
based CNN）Network, the test average detected precision on VOC（mean Average Precision, mAP）It is
Felzenszwalb et al. proposes the DPM based on HOG（Deformable Parts Model）Twice of algorithm of target detection.From
After R-CNN appearance, the performance based on the target detection of CNN in VOC data sets occupies leading position, and it is big to be broadly divided into two
Class：（1）Target detection based on candidate region, its main feature is that with very high accuracy of detection, but speed cannot meet application
Requirement of real-time, wherein masterpiece have the R-FCN, Faster R-CNN of 2017, the Mask R-CNN of 2017 of 2016
Deng；（2）Target detection based on recurrence, its main feature is that speed is quickly, but accuracy of detection is not good enough, and masterpiece has 2015
YOLO（You Only Look Once）With the SSD of 2016（Single Shot Multibox Detector）Deng.
Jonathan Huang et al. illustrated meta structure in 2016（SSD, Faster R-CNN and R-FCN）Accuracy of detection
The method compromised between speed.In addition to this, some cascade method for detecting human face also have good effect, for example,
The Joint Cascade methods that Chen et al. was proposed in 2014 carry out grade using the reference points detection of Face datection and face
Connection has higher detection result in traditional method for detecting human face；The MTCNN networks that Zhang et al. was proposed in 2016
It is cascaded using three convolutional networks, the algorithm structure of " from slightly to essence " causes being detected to face for multitask, has higher
Recall rate, but need to use three kinds of different data sets during training network, it is relatively complicated；What Yang et al. was proposed in 2016
Faceness networks judge whether detected target is face using nose, face, eyes, hair, this five features of beard,
With higher accuracy of detection, but it is unsatisfactory for real-time criterion.Deep learning develops to embedded devices such as mobile phones, is
Reach real-time demand, there is very high limitation, therefore Andrew G. Howard in 2017 to the number of parameters of basic network
Et al. propose MobileNet, it exchanges a large amount of parameter for a small amount of nicety of grading and reduces.The number of parameters of MoblieNet is
The 1/33 of VGG16, ImageNet-1000 classification accuracy for 70.6%, only fewer than VGG16 0.9%.In conclusion at present
It is still a big difficulty that speed and precision are taken into account in object detection field.
Invention content
In practical engineering application, most of is the face in image sequence to be detected, and system requirements is steady in real time
Surely angle changed greatly, block more serious face and be detected.Therefore the quick MobileNet bases utilized herein
Plinth network is improved and is combined to form MS with quick Face datection network SSD models（MobileNet-SSD）Detect net
Network can be good at taking into account detection speed and precision, MS network parameters is adjusted herein, complies with two classification（Face
Target and background）Face datection task, recycle core correlation filtering（Kernelized Correlation Filters,
KCF）Algorithm carries out the face detected stable tracking, forms detection-tracking-detection（DTD）Pattern is cyclically updated, i.e.,
MS-KCF Face datection models.The model not only solves that angle change is larger, blocks more serious Face datection stability
Problem, and the detection speed of human face target in image sequence can be greatly improved.
The technology bill of the present invention is as follows：Face fast and stable detection method in a kind of image sequence based on MS-KCF,
It is main to include step：
Step 1, MS is built（MobileNet-SSD）Detect network；
Step 2, image sequence is read, image is detected using MS detection networks；
Step 3, trace model is updated, the coordinate information for detecting human face target is passed into KCF trackers, as tracking
The basic sample pane of device, and specimen sample and training are nearby carried out to sample pane, for predicting the position of next frame human face target；
Step 4, the phenomenon that human face target is lost when tracking in order to prevent after tracking number frame, then updates MS detection networks, to people
Face target detects positioning again；
Step 5, experimental result is made comparisons analysis with currently advanced method for detecting human face.
Description of the drawings
Fig. 1 is the system overview flow chart of the present invention
Fig. 2 is the MS network structures of the present invention
Fig. 3 is the improved MobileNet convolutional coding structures figure of the present invention
Fig. 4 is the MS network convolution feature pyramid diagrams of the present invention
Fig. 5 is the test result figure of the MS-KCF models of the present invention
Fig. 6 is the ROC curve comparison diagram of the Girl image sequences of the present invention
Fig. 7 is the ROC curve comparison diagram of the FaceOcc1 image sequences of the present invention.
Specific embodiment
Below by the present invention the image sequence based on MS-KCF in face fast and stable detection method combination example with
Attached drawing is described in further detail.
As shown in Figure 1, the system overview flow chart for the present invention, comprising image sequence module is obtained, MS detects network mould
Block, KCF tracking modules, model modification module.So as to which whole network forms a kind of new automatic detection-tracking-detection
（Detection-Tracking-Detection, DTD）It is cyclically updated pattern, i.e. MS-KCF Face datections model.
Step 1, MS is built（MobileNet-SSD）Detect network.As shown in Fig. 2, MS detection network structures include four
Part：First part is input layer, for inputting picture；Second part is improved MobileNet convolutional networks, for extracting
Input the feature of picture；Part III is SSD meta structures, is returned for classifying to return with bounding box；Part IV is output layer,
For exporting testing result.As shown in table 1, the general frame of network is detected for MS, Conv_BN_ReLU6 represents Standard convolution
Layer, Conv1_Dw_Pw represent that depth separates convolutional layer, and ' √ ' represents the characteristic pattern of convolutional layer output, it will for classifying
It returns in being returned with bounding box.Since human face target is smaller, therefore the feature of the Conv7_Dw_Pw outputs compared with shallow-layer has been taken herein
Figure.
1 MS general frames of table
MS detection networks include improved MobileNet convolutional networks and SSD meta structure two parts.
（1）Improved MobileNet convolutional networks extract feature.As shown in figure 3, for improved MobileNet convolution knot
Structure：Conv_Dw_Pw is the separable convolution of depth（Depthwise Separable Convolutions）, Dw is the depth of 3x3
Layer convolutional layer（Depthwise Layers）, Pw is the point convolutional layer of 1x1（Pointwise Layers）, and each convolution
Operation all normalizes below by batch（Batch Normalization, BN）Algorithm and activation primitive ReLU6.The present invention will
Activation primitive ReLU in MoblieNet networks is changed to ReLU6, and the BN algorithms of cooperation automatic adjustment data distribution accelerate instruction
Experienced convergence rate.Formula (1) is ReLU6 activation primitives：
(1)
WhereinxIt is the input of activation primitive,yIt is output.
Improved MobileNet convolutional coding structures are designed very ingenious.
First, the separable convolutional coding structure of depth greatly reduced calculation amount, and convergent speed when accelerating trained
Degree, the reason is as follows that：
In the calculating for carrying out Standard convolution, if the size of input picture is,MRepresent the port number of input picture,N
Represent the port number of convolution output, Standard convolution core size is.If it calculates cost to be represented with number of parameters, this side
Method calculates cost：。
But for depth in MobileNet separate convolutional calculation formula and it is above-mentioned similarly output and input, preceding
Half portion Dw stages, the size of depth convolution kernel needed are, in the latter half of point convolution kernel in the Pw stages, needed
Size is, the calculating cost that depth separates convolution at this time is：,
It is that Standard convolution calculates costTimes.
Second, convolutional neural networks train during, due to each layer of convolution all can change data distribution.If
Data distribution is at the edge of activation primitive, it will gradient is caused to disappear so that parameter no longer updates.BN algorithms are by setting two
The parameter that can learn adjusts the distribution of data（Similar to standardized normal distribution）, avoid the gradient disappearance in training process
Phenomenon and complicated parameter（Learning rate, Dropout ratios etc.）Setting.
（2）SSD meta structures return.SSD networks are a kind of regression models, and the feature exported using different convolutional layers is divided
Class returns and bounding box returns, and not only preferably alleviates translation invariance and translates the contradiction between changeability, but also to inspection
Surveying accuracy and speed has a preferable compromise, i.e., with higher accuracy of detection while detection speed is improved.For 300*
The input of 300 sizes, under Titan X GPU hardware environment test voc2007 data sets, with 59fps detection speed and
74.3% average detected precision（mean Average Precision, mAP）.SSD is a training pattern end to end,
Overall loss function during trainingLThe position loss that the confidence level loss and bounding box returned including classification returns, is defined as：
(2)
In formula (2)xRepresent the feature of input；cPresentation class confidence level；lThe offset of prediction is represented, including center point coordinate
Translational offsets and boundary frame width it is high scaling offset；gCalibration frame for target actual positions；It is put for what classification returned
Reliability is lost；The position loss returned for bounding box；It is the parameter for balancing two kinds of losses；PRepresent matching
Acquiescence frame quantity, whenPWhen being 0, overall loss will be arranged to 0.
Step 2, image sequence is read, image is detected using MS detection networks.It is illustrated in figure 4 MS network convolution
Feature pyramid, in order to meet the translation changeability of Detection task requirement, the present invention is obtained respectively two in improved MobileNet
Four layers of characteristic pattern composition characteristic figure pyramid in layer characteristic pattern and additional Standard convolution layer, then the convolution with different 3*3
Core carries out convolution, and the result after convolution carries out that classification returns and bounding box is returned as final feature.The present invention is big with 300*300
Small picture as input, in above-mentioned six layers of convolution characteristic pattern pyramid the acquiescence frame number of each feature unit be respectively 4,
6、6、6、6、6.And 3x3 sizes, the convolution nuclear parameter that step-length is 1 used in different layers, different task are different from.
Step 3, trace model is updated, the coordinate information for detecting human face target is passed into KCF trackers, as
The basic sample pane of tracker, and specimen sample and training are nearby carried out to sample pane, for predicting next frame human face target
Position.
The face that is moved in image sequence is angled to be changed greatly, blocks the problems such as more serious, face can be caused to examine
During survey the phenomenon that missing inspection.KCF is quick target tracking algorism, therefore model modification is carried out during Face datection：Profit
Detect that starting KCF algorithms while face carries out continual and steady tracking with MS detection networks, after 10 frames are tracked, in order to keep away
Exempt from the loss of tracking, again with face detection model more new target location.Therefore, KCF algorithms play the role of be：
（1）Face datection is to the robustness of the variations such as posture, angle in reinforcement image sequence；
（2）Play the role of linking and acceleration in DTD models, substantially increase the detection speed of whole system.
IfTo input,For label, then training sample set is, number of samples isR, the purpose of recurrence is to pass through
Find a kind of mapping relationsfSo that, linear regression function is, whereinRepresent weight coefficient.
Formula (3) is error function used in the algorithm：
(3)
Wherein coefficientFor the structural complexity of control system, to ensure the generalization ability of grader.Formula (3) is carried out minimum
Square law solves to obtain optimal weight coefficientw：
(4)
In formula (4)。TRepresent transposition,IRepresent unit matrix,XIn per a line
Represent a feature vector.Formula (5) is the complex field form of formula (4)：
(5)
WhereinIt representsXComplex conjugate transposed matrix.It solves at this timewCalculating time complexity be。
In KCF algorithms, training sample and test sample are all by basic sampleWhat is generated follows
What ring matrix was formed, i.e.,：
(6)
In formula (6)The discrete fourier matrix in formula (7) can be passed throughFIt obtains：
(7)
(8)
(9)
(10)
In formula (8),For basic sampleDiscrete fourier version,It representsFComplex conjugate transposed matrix.Formula (9)
In,ForHermitian transposition, "diag" it is diagonalization of matrix operation.Formula (10) is the deformation of formula (9), wherein "" be
It is operated by the multiplication of element.After being first carried out at the same time discrete Fourier transform to formula (5) both sides, further in accordance with formula (8 ~ 10), knot is obtained
Fruit is：
(11)
In formula (11),ForYDiscrete Fourier transform, forIt is i.e. available that Fourier inversion is carried out againw.This up-to-date style
(11) inwThe calculating time complexity of solution isO(n), the time complexity of discrete Fourier transform isO(nlogn), compared to it
BeforewCalculating time complexityGreatly reduce the time complexity of whole system.
KCF algorithm objectives are the circular matrix by Fourier space to reduce the calculating time complexity of recurrence, from
And it obtains a large amount of speed and is promoted.
Step 4, the phenomenon that human face target is lost when tracking in order to prevent after tracking 10 frames, then updates MS detection models, right
Human face target detects positioning again, so that whole network forms a kind of new automatic detection-tracking-detection
（Detection-Tracking-Detection, DTD）Pattern, i.e. MS-KCF Face datections model so that entire detection process
Speed and precision are taken into account.
Step 5, experimental result is made comparisons analysis with currently advanced method for detecting human face.
Test is assessed on GTX1080 GPU, and input picture is scaled the size of 300*300.
Table 2 is the FDDB Static Human Faces detection data concentration average recall rate of distinct methods and average speed pair in standard
Compare result.Table 2 shows that this method has preferable recall rate, and MS detects 2.8 times faster than MTCNN of the detection speed of network, than
Fast 9.3 times of Faceness.Therefore, this method has higher recall rate and quickly detection on Static Human Face Test database
Speed.
Table 2 is averaged recall rate and average speed in FDDB data sets with distinct methods
Fig. 5 (a) and Fig. 5 (b) is two image sequences concentrated in VOT2016 dynamic human faces tracking data respectively（Girl and
FaceOcc1）Test result.Girl is the image sequence that facial angle changes greatly, and FaceOcc1 is to block larger image
Sequence.Preceding two rows of testing results for MS models of each image sequence in Fig. 5 (a) and Fig. 5 (b), rear two rows are MS-KCF
The testing result of model.Obviously, MS-KCF models for angle change in image sequence it is larger, block more serious face and have
There is better detection performance.
Fig. 6 and Fig. 7 is the ROC curve of two image sequences Girl and FaceOcc1 in VOT2016 data sets respectively
Comparing result.By Fig. 6 and Fig. 7 it is found that for the Face datection task in image sequence, there is the MS-KCF of model modification
The detection performance of method is better than the MS methods only with detection function.
Table 3 in VOT2016 data sets with distinct methods average speed
Table 3 is the average speed comparing result of the distinct methods in VOT2016 data sets.As shown in Table 3, there is model modification work(
The MS-KCF methods of energy are quick, and detection speed is 2.3 times faster than the MS methods only with detection function, faster than MTCNN 6.4
Times, 21.4 times faster than Faceness.
Claims (4)
1. a kind of face fast and stable detection method in image sequence based on MS-KCF, including following five steps：
Step 1, MS is built（MobileNet-SSD）Detect network；
Step 2, image sequence is read, image is detected using MS networks；
Step 3, trace model is updated, the coordinate information for detecting human face target is passed into core correlation filtering（Kernelized
Correlation Filters, KCF）Tracker is nearby carried out as the basic sample pane of tracker, and to sample pane
Specimen sample and training, for predicting the position of next frame human face target；
Step 4, the phenomenon that human face target is lost when tracking in order to prevent after tracking number frame, then updates MS detection networks, to people
Face target detects positioning again；
Step 5, experimental result is made comparisons analysis with currently advanced method for detecting human face.
2. face fast and stable detection method in a kind of image sequence based on MS-KCF according to claim 1, special
Sign is that MS detects network and employs improved quick accurate MobileNet networks to replace in former SSD models in step 1
Baseline network VGG, original MobileNet in Pw structures change the distributions of Dw structure output data so that it detects essence
Degree declines, therefore the present invention has cast out the full articulamentum of original MobileNet, additionally increases by 8 layers of Standard convolution layer, for expanding
The receptive field of characteristic pattern, adjustment data distribution and the translation invariance for strengthening classification task requirement, gradient disappearance in order to prevent,
It is normalized behind each layer of convolutional layer plus batch（Batch Normalization, BN）Layer, activation primitive are changed to by ReLU
ReLU6。
3. face fast and stable detection method in a kind of image sequence based on MS-KCF according to claim 1, special
Sign is that MS proposed by the invention detects network to meet the translation changeability of Detection task requirement in step 2, obtains respectively
Take four layers of characteristic pattern composition characteristic figure gold word in two layers of characteristic pattern and additional Standard convolution layer in improved MobileNet
Tower, then carry out convolution with the convolution kernel of different 3*3, the result after convolution carries out classification recurrence and bounding box as final feature
It returns.
4. face fast and stable detection method in a kind of image sequence based on MS-KCF according to claim 1, special
Sign is that the model modification twice in step 3 and step 4 realizes the accurate detection positioning to human face target, so that whole
A network forms a kind of new automatic detection-tracking-detection（Detection-Tracking-Detection, DTD）It follows
Ring renewal model, i.e. MS-KCF Face datections model so that entire detection process both ensure that angle change in image sequence
It is larger, block the stability of more serious Face datection, while also substantially increase detection speed.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810124952.2A CN108229442B (en) | 2018-02-07 | 2018-02-07 | Method for rapidly and stably detecting human face in image sequence based on MS-KCF |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810124952.2A CN108229442B (en) | 2018-02-07 | 2018-02-07 | Method for rapidly and stably detecting human face in image sequence based on MS-KCF |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108229442A true CN108229442A (en) | 2018-06-29 |
| CN108229442B CN108229442B (en) | 2022-03-11 |
Family
ID=62671130
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810124952.2A Active CN108229442B (en) | 2018-02-07 | 2018-02-07 | Method for rapidly and stably detecting human face in image sequence based on MS-KCF |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108229442B (en) |
Cited By (49)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109034119A (en) * | 2018-08-27 | 2018-12-18 | 苏州广目信息技术有限公司 | A kind of method for detecting human face of the full convolutional neural networks based on optimization |
| CN109063666A (en) * | 2018-08-14 | 2018-12-21 | 电子科技大学 | The lightweight face identification method and system of convolution are separated based on depth |
| CN109086678A (en) * | 2018-07-09 | 2018-12-25 | 天津大学 | A kind of pedestrian detection method extracting image multi-stage characteristics based on depth supervised learning |
| CN109118519A (en) * | 2018-07-26 | 2019-01-01 | 北京纵目安驰智能科技有限公司 | Target Re-ID method, system, terminal and the storage medium of Case-based Reasoning segmentation |
| CN109146967A (en) * | 2018-07-09 | 2019-01-04 | 上海斐讯数据通信技术有限公司 | The localization method and device of target object in image |
| CN109145836A (en) * | 2018-08-28 | 2019-01-04 | 武汉大学 | Ship target video detection method based on deep learning network and Kalman filtering |
| CN109271848A (en) * | 2018-08-01 | 2019-01-25 | 深圳市天阿智能科技有限责任公司 | A kind of method for detecting human face and human face detection device, storage medium |
| CN109344731A (en) * | 2018-09-10 | 2019-02-15 | 电子科技大学 | The face identification method of lightweight neural network based |
| CN109409210A (en) * | 2018-09-11 | 2019-03-01 | 北京飞搜科技有限公司 | A kind of method for detecting human face and system based on SSD frame |
| CN109472315A (en) * | 2018-11-15 | 2019-03-15 | 江苏木盟智能科技有限公司 | A kind of object detection method and system separating convolution based on depth |
| CN109492674A (en) * | 2018-10-19 | 2019-03-19 | 北京京东尚科信息技术有限公司 | The generation method and device of SSD frame for target detection |
| CN109523476A (en) * | 2018-11-02 | 2019-03-26 | 武汉烽火众智数字技术有限责任公司 | License plate for video investigation goes motion blur method |
| CN109558877A (en) * | 2018-10-19 | 2019-04-02 | 复旦大学 | Naval target track algorithm based on KCF |
| CN109583443A (en) * | 2018-11-15 | 2019-04-05 | 四川长虹电器股份有限公司 | A kind of video content judgment method based on Text region |
| CN109598742A (en) * | 2018-11-27 | 2019-04-09 | 湖北经济学院 | A kind of method for tracking target and system based on SSD algorithm |
| CN109711332A (en) * | 2018-12-26 | 2019-05-03 | 浙江捷尚视觉科技股份有限公司 | A kind of face tracking method and application based on regression algorithm |
| CN109754071A (en) * | 2018-12-29 | 2019-05-14 | 北京中科寒武纪科技有限公司 | Activate operation method, device, electronic equipment and readable storage medium storing program for executing |
| CN109828251A (en) * | 2019-03-07 | 2019-05-31 | 中国人民解放军海军航空大学 | Radar target identification method based on feature pyramid light weight convolutional neural networks |
| CN109840502A (en) * | 2019-01-31 | 2019-06-04 | 深兰科技（上海）有限公司 | A kind of method and apparatus carrying out target detection based on SSD model |
| CN109903507A (en) * | 2019-03-04 | 2019-06-18 | 上海海事大学 | A kind of fire disaster intelligent monitor system and method based on deep learning |
| CN109978045A (en) * | 2019-03-20 | 2019-07-05 | 深圳市道通智能航空技术有限公司 | A kind of method for tracking target, device and unmanned plane |
| CN109993052A (en) * | 2018-12-26 | 2019-07-09 | 上海航天控制技术研究所 | The method for tracking target and system of dimension self-adaption under a kind of complex scene |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN110287849A (en) * | 2019-06-20 | 2019-09-27 | 北京工业大学 | A kind of lightweight depth network image object detection method suitable for raspberry pie |
| CN110298225A (en) * | 2019-03-28 | 2019-10-01 | 电子科技大学 | A method of blocking the human face five-sense-organ positioning under environment |
| CN110348423A (en) * | 2019-07-19 | 2019-10-18 | 西安电子科技大学 | A kind of real-time face detection method based on deep learning |
| CN110363137A (en) * | 2019-07-12 | 2019-10-22 | 创新奇智(广州)科技有限公司 | Face datection Optimized model, method, system and its electronic equipment |
| CN110378239A (en) * | 2019-06-25 | 2019-10-25 | 江苏大学 | A kind of real-time traffic marker detection method based on deep learning |
| CN110414371A (en) * | 2019-07-08 | 2019-11-05 | 西南科技大学 | A kind of real-time face expression recognition method based on multiple dimensioned nuclear convolution neural network |
| CN110443247A (en) * | 2019-08-22 | 2019-11-12 | 中国科学院国家空间科学中心 | A kind of unmanned aerial vehicle moving small target real-time detecting system and method |
| CN110490899A (en) * | 2019-07-11 | 2019-11-22 | 东南大学 | A kind of real-time detection method of the deformable construction machinery of combining target tracking |
| CN110495962A (en) * | 2019-08-26 | 2019-11-26 | 赫比（上海）家用电器产品有限公司 | The method and its toothbrush and equipment of monitoring toothbrush position |
| CN110580445A (en) * | 2019-07-12 | 2019-12-17 | 西北工业大学 | Face key point detection method based on GIoU and weighted NMS improvement |
| CN110619279A (en) * | 2019-08-22 | 2019-12-27 | 天津大学 | Road traffic sign instance segmentation method based on tracking |
| CN110647810A (en) * | 2019-08-16 | 2020-01-03 | 西北大学 | Method and device for constructing and identifying radio signal image identification model |
| CN110909688A (en) * | 2019-11-26 | 2020-03-24 | 南京甄视智能科技有限公司 | Face detection small model optimization training method, face detection method and computer system |
| CN110956082A (en) * | 2019-10-17 | 2020-04-03 | 江苏科技大学 | Face key point detection method and detection system based on deep learning |
| CN111104817A (en) * | 2018-10-25 | 2020-05-05 | 中车株洲电力机车研究所有限公司 | Fatigue detection method based on deep learning |
| CN111160269A (en) * | 2019-12-30 | 2020-05-15 | 广东工业大学 | Face key point detection method and device |
| CN111325157A (en) * | 2020-02-24 | 2020-06-23 | 高新兴科技集团股份有限公司 | Face snapshot method, computer storage medium and electronic device |
| CN111339858A (en) * | 2020-02-17 | 2020-06-26 | 电子科技大学 | Oil and gas pipeline marker identification method based on neural network |
| CN111339832A (en) * | 2020-02-03 | 2020-06-26 | 中国人民解放军国防科技大学 | Method and device for detecting face synthetic image |
| CN111667505A (en) * | 2020-04-30 | 2020-09-15 | 北京捷通华声科技股份有限公司 | Method and device for tracking fixed object |
| CN111814827A (en) * | 2020-06-08 | 2020-10-23 | 三峡大学 | Key point target detection method based on YOLO |
| CN111860046A (en) * | 2019-04-26 | 2020-10-30 | 四川大学 | Facial expression recognition method for improving MobileNet model |
| CN112487852A (en) * | 2019-09-12 | 2021-03-12 | 上海齐感电子信息科技有限公司 | Face detection method and device for embedded equipment, storage medium and terminal |
| CN112581506A (en) * | 2020-12-31 | 2021-03-30 | 北京澎思科技有限公司 | Face tracking method, system and computer readable storage medium |
| CN112801117A (en) * | 2021-02-03 | 2021-05-14 | 四川中烟工业有限责任公司 | Multi-channel receptive field guided characteristic pyramid small target detection network and detection method |
| US11281923B2 (en) | 2019-02-19 | 2022-03-22 | Fujitsu Limited | Object recognition method, apparatus and network |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106127776A (en) * | 2016-06-28 | 2016-11-16 | 北京工业大学 | Based on multiple features space-time context robot target identification and motion decision method |
| CN106204638A (en) * | 2016-06-29 | 2016-12-07 | 西安电子科技大学 | A kind of based on dimension self-adaption with the method for tracking target of taking photo by plane blocking process |
| WO2017088050A1 (en) * | 2015-11-26 | 2017-06-01 | Sportlogiq Inc. | Systems and methods for object tracking and localization in videos with adaptive image representation |
| CN106960446A (en) * | 2017-04-01 | 2017-07-18 | 广东华中科技大学工业技术研究院 | A kind of waterborne target detecting and tracking integral method applied towards unmanned boat |
| CN107066953A (en) * | 2017-03-22 | 2017-08-18 | 北京邮电大学 | It is a kind of towards the vehicle cab recognition of monitor video, tracking and antidote and device |
-
2018
- 2018-02-07 CN CN201810124952.2A patent/CN108229442B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2017088050A1 (en) * | 2015-11-26 | 2017-06-01 | Sportlogiq Inc. | Systems and methods for object tracking and localization in videos with adaptive image representation |
| CN106127776A (en) * | 2016-06-28 | 2016-11-16 | 北京工业大学 | Based on multiple features space-time context robot target identification and motion decision method |
| CN106204638A (en) * | 2016-06-29 | 2016-12-07 | 西安电子科技大学 | A kind of based on dimension self-adaption with the method for tracking target of taking photo by plane blocking process |
| CN107066953A (en) * | 2017-03-22 | 2017-08-18 | 北京邮电大学 | It is a kind of towards the vehicle cab recognition of monitor video, tracking and antidote and device |
| CN106960446A (en) * | 2017-04-01 | 2017-07-18 | 广东华中科技大学工业技术研究院 | A kind of waterborne target detecting and tracking integral method applied towards unmanned boat |
Non-Patent Citations (3)
| Title |
|---|
| J. F. HENRIQUES: "High-speed tracking with kernelized correlation filters", 《 IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE》 * |
| 叶远征等: "基于MS-KCF模型的图像序列中人脸快速稳定检测", 《计算机应用》 * |
| 王杨: "基于深度信息的核相关性滤波跟踪算法研究", 《中国优秀博硕士学位论文全文数据库(硕士)信息科技辑》 * |
Cited By (70)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109086678A (en) * | 2018-07-09 | 2018-12-25 | 天津大学 | A kind of pedestrian detection method extracting image multi-stage characteristics based on depth supervised learning |
| CN109146967A (en) * | 2018-07-09 | 2019-01-04 | 上海斐讯数据通信技术有限公司 | The localization method and device of target object in image |
| CN109086678B (en) * | 2018-07-09 | 2022-02-25 | 天津大学 | Pedestrian detection method for extracting image multilevel features based on deep supervised learning |
| CN109118519A (en) * | 2018-07-26 | 2019-01-01 | 北京纵目安驰智能科技有限公司 | Target Re-ID method, system, terminal and the storage medium of Case-based Reasoning segmentation |
| CN109271848B (en) * | 2018-08-01 | 2022-04-15 | 深圳市天阿智能科技有限责任公司 | Face detection method, face detection device and storage medium |
| CN109271848A (en) * | 2018-08-01 | 2019-01-25 | 深圳市天阿智能科技有限责任公司 | A kind of method for detecting human face and human face detection device, storage medium |
| CN109063666A (en) * | 2018-08-14 | 2018-12-21 | 电子科技大学 | The lightweight face identification method and system of convolution are separated based on depth |
| CN109034119A (en) * | 2018-08-27 | 2018-12-18 | 苏州广目信息技术有限公司 | A kind of method for detecting human face of the full convolutional neural networks based on optimization |
| CN109145836A (en) * | 2018-08-28 | 2019-01-04 | 武汉大学 | Ship target video detection method based on deep learning network and Kalman filtering |
| CN109145836B (en) * | 2018-08-28 | 2021-04-16 | 武汉大学 | Ship target video detection method based on deep learning network and Kalman filtering |
| CN109344731B (en) * | 2018-09-10 | 2022-05-03 | 电子科技大学 | Lightweight face recognition method based on neural network |
| CN109344731A (en) * | 2018-09-10 | 2019-02-15 | 电子科技大学 | The face identification method of lightweight neural network based |
| CN109409210B (en) * | 2018-09-11 | 2020-11-24 | 苏州飞搜科技有限公司 | Face detection method and system based on SSD (solid State disk) framework |
| CN109409210A (en) * | 2018-09-11 | 2019-03-01 | 北京飞搜科技有限公司 | A kind of method for detecting human face and system based on SSD frame |
| CN109492674B (en) * | 2018-10-19 | 2020-11-03 | 北京京东尚科信息技术有限公司 | Generation method and device of SSD (solid State disk) framework for target detection |
| CN109558877B (en) * | 2018-10-19 | 2023-03-07 | 复旦大学 | KCF-based offshore target tracking algorithm |
| CN109558877A (en) * | 2018-10-19 | 2019-04-02 | 复旦大学 | Naval target track algorithm based on KCF |
| CN109492674A (en) * | 2018-10-19 | 2019-03-19 | 北京京东尚科信息技术有限公司 | The generation method and device of SSD frame for target detection |
| CN111104817A (en) * | 2018-10-25 | 2020-05-05 | 中车株洲电力机车研究所有限公司 | Fatigue detection method based on deep learning |
| CN109523476A (en) * | 2018-11-02 | 2019-03-26 | 武汉烽火众智数字技术有限责任公司 | License plate for video investigation goes motion blur method |
| CN109523476B (en) * | 2018-11-02 | 2022-04-05 | 武汉烽火众智数字技术有限责任公司 | License plate motion blur removing method for video detection |
| CN109583443A (en) * | 2018-11-15 | 2019-04-05 | 四川长虹电器股份有限公司 | A kind of video content judgment method based on Text region |
| CN109472315B (en) * | 2018-11-15 | 2021-09-24 | 江苏木盟智能科技有限公司 | Target detection method and system based on depth separable convolution |
| CN109472315A (en) * | 2018-11-15 | 2019-03-15 | 江苏木盟智能科技有限公司 | A kind of object detection method and system separating convolution based on depth |
| CN109598742A (en) * | 2018-11-27 | 2019-04-09 | 湖北经济学院 | A kind of method for tracking target and system based on SSD algorithm |
| CN109993052A (en) * | 2018-12-26 | 2019-07-09 | 上海航天控制技术研究所 | The method for tracking target and system of dimension self-adaption under a kind of complex scene |
| CN109993052B (en) * | 2018-12-26 | 2021-04-13 | 上海航天控制技术研究所 | Scale-adaptive target tracking method and system under complex scene |
| CN109711332B (en) * | 2018-12-26 | 2021-03-26 | 浙江捷尚视觉科技股份有限公司 | Regression algorithm-based face tracking method and application |
| CN109711332A (en) * | 2018-12-26 | 2019-05-03 | 浙江捷尚视觉科技股份有限公司 | A kind of face tracking method and application based on regression algorithm |
| CN109754071A (en) * | 2018-12-29 | 2019-05-14 | 北京中科寒武纪科技有限公司 | Activate operation method, device, electronic equipment and readable storage medium storing program for executing |
| CN109754071B (en) * | 2018-12-29 | 2020-05-05 | 中科寒武纪科技股份有限公司 | Activation operation method and device, electronic equipment and readable storage medium |
| CN109840502A (en) * | 2019-01-31 | 2019-06-04 | 深兰科技（上海）有限公司 | A kind of method and apparatus carrying out target detection based on SSD model |
| CN109840502B (en) * | 2019-01-31 | 2021-06-15 | 深兰科技（上海）有限公司 | Method and device for target detection based on SSD model |
| US11281923B2 (en) | 2019-02-19 | 2022-03-22 | Fujitsu Limited | Object recognition method, apparatus and network |
| CN109903507A (en) * | 2019-03-04 | 2019-06-18 | 上海海事大学 | A kind of fire disaster intelligent monitor system and method based on deep learning |
| CN109828251A (en) * | 2019-03-07 | 2019-05-31 | 中国人民解放军海军航空大学 | Radar target identification method based on feature pyramid light weight convolutional neural networks |
| CN109978045A (en) * | 2019-03-20 | 2019-07-05 | 深圳市道通智能航空技术有限公司 | A kind of method for tracking target, device and unmanned plane |
| CN110009015A (en) * | 2019-03-25 | 2019-07-12 | 西北工业大学 | EO-1 hyperion small sample classification method based on lightweight network and semi-supervised clustering |
| CN110298225A (en) * | 2019-03-28 | 2019-10-01 | 电子科技大学 | A method of blocking the human face five-sense-organ positioning under environment |
| CN111860046A (en) * | 2019-04-26 | 2020-10-30 | 四川大学 | Facial expression recognition method for improving MobileNet model |
| CN111860046B (en) * | 2019-04-26 | 2022-10-11 | 四川大学 | Facial expression recognition method for improving MobileNet model |
| CN110287849A (en) * | 2019-06-20 | 2019-09-27 | 北京工业大学 | A kind of lightweight depth network image object detection method suitable for raspberry pie |
| CN110378239A (en) * | 2019-06-25 | 2019-10-25 | 江苏大学 | A kind of real-time traffic marker detection method based on deep learning |
| CN110414371A (en) * | 2019-07-08 | 2019-11-05 | 西南科技大学 | A kind of real-time face expression recognition method based on multiple dimensioned nuclear convolution neural network |
| CN110490899A (en) * | 2019-07-11 | 2019-11-22 | 东南大学 | A kind of real-time detection method of the deformable construction machinery of combining target tracking |
| CN110580445B (en) * | 2019-07-12 | 2023-02-07 | 西北工业大学 | Face key point detection method based on GIoU and weighted NMS improvement |
| CN110580445A (en) * | 2019-07-12 | 2019-12-17 | 西北工业大学 | Face key point detection method based on GIoU and weighted NMS improvement |
| CN110363137A (en) * | 2019-07-12 | 2019-10-22 | 创新奇智(广州)科技有限公司 | Face datection Optimized model, method, system and its electronic equipment |
| CN110348423A (en) * | 2019-07-19 | 2019-10-18 | 西安电子科技大学 | A kind of real-time face detection method based on deep learning |
| CN110647810A (en) * | 2019-08-16 | 2020-01-03 | 西北大学 | Method and device for constructing and identifying radio signal image identification model |
| CN110443247A (en) * | 2019-08-22 | 2019-11-12 | 中国科学院国家空间科学中心 | A kind of unmanned aerial vehicle moving small target real-time detecting system and method |
| CN110619279A (en) * | 2019-08-22 | 2019-12-27 | 天津大学 | Road traffic sign instance segmentation method based on tracking |
| CN110495962A (en) * | 2019-08-26 | 2019-11-26 | 赫比（上海）家用电器产品有限公司 | The method and its toothbrush and equipment of monitoring toothbrush position |
| CN112487852A (en) * | 2019-09-12 | 2021-03-12 | 上海齐感电子信息科技有限公司 | Face detection method and device for embedded equipment, storage medium and terminal |
| CN110956082B (en) * | 2019-10-17 | 2023-03-24 | 江苏科技大学 | Face key point detection method and detection system based on deep learning |
| CN110956082A (en) * | 2019-10-17 | 2020-04-03 | 江苏科技大学 | Face key point detection method and detection system based on deep learning |
| CN110909688A (en) * | 2019-11-26 | 2020-03-24 | 南京甄视智能科技有限公司 | Face detection small model optimization training method, face detection method and computer system |
| CN111160269A (en) * | 2019-12-30 | 2020-05-15 | 广东工业大学 | Face key point detection method and device |
| CN111339832A (en) * | 2020-02-03 | 2020-06-26 | 中国人民解放军国防科技大学 | Method and device for detecting face synthetic image |
| CN111339832B (en) * | 2020-02-03 | 2023-09-12 | 中国人民解放军国防科技大学 | Face synthetic image detection method and device |
| CN111339858A (en) * | 2020-02-17 | 2020-06-26 | 电子科技大学 | Oil and gas pipeline marker identification method based on neural network |
| CN111339858B (en) * | 2020-02-17 | 2022-07-29 | 电子科技大学 | Oil and gas pipeline marker identification method based on neural network |
| CN111325157A (en) * | 2020-02-24 | 2020-06-23 | 高新兴科技集团股份有限公司 | Face snapshot method, computer storage medium and electronic device |
| CN111667505A (en) * | 2020-04-30 | 2020-09-15 | 北京捷通华声科技股份有限公司 | Method and device for tracking fixed object |
| CN111667505B (en) * | 2020-04-30 | 2023-04-07 | 北京捷通华声科技股份有限公司 | Method and device for tracking fixed object |
| CN111814827A (en) * | 2020-06-08 | 2020-10-23 | 三峡大学 | Key point target detection method based on YOLO |
| CN111814827B (en) * | 2020-06-08 | 2024-06-11 | 湖南腓腓动漫有限责任公司 | YOLO-based key point target detection method |
| CN112581506A (en) * | 2020-12-31 | 2021-03-30 | 北京澎思科技有限公司 | Face tracking method, system and computer readable storage medium |
| CN112801117B (en) * | 2021-02-03 | 2022-07-12 | 四川中烟工业有限责任公司 | Multi-channel receptive field guided characteristic pyramid small target detection network and detection method |
| CN112801117A (en) * | 2021-02-03 | 2021-05-14 | 四川中烟工业有限责任公司 | Multi-channel receptive field guided characteristic pyramid small target detection network and detection method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108229442B (en) | 2022-03-11 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108229442A (en) | Face fast and stable detection method in image sequence based on MS-KCF | |
| CN109800689B (en) | Target tracking method based on space-time feature fusion learning | |
| CN108399373B (en) | The model training and its detection method and device of face key point | |
| CN111882580A (en) | Video multi-target tracking method and system | |
| CN112801169B (en) | Camouflage target detection method, system, device and storage medium based on improved YOLO algorithm | |
| CN110533691B (en) | Target tracking method, device and storage medium based on multiple classifiers | |
| Chetverikov et al. | Dynamic texture as foreground and background | |
| CN110097091A (en) | It is trained be distributed with inference data it is inconsistent under the conditions of image fine granularity recognition methods | |
| CN110263731B (en) | Single step human face detection system | |
| CN116229192B (en) | ODConvBS-YOLOv s-based flame smoke detection method | |
| CN105608710A (en) | Non-rigid face detection and tracking positioning method | |
| CN116546617A (en) | Ray tracing fingerprint positioning method and device based on non-vision scene | |
| CN113420289B (en) | Hidden poisoning attack defense method and device for deep learning model | |
| CN105335703B (en) | Traffic scene method for detecting abnormality based on motor reorgan technology | |
| CN117237547B (en) | Image reconstruction method, reconstruction model processing method and device | |
| CN113763440A (en) | Image processing method, device, equipment and storage medium | |
| CN116403150A (en) | Mask detection algorithm based on C3-CBAM (C3-CBAM) attention mechanism | |
| CN113450321B (en) | Single-stage target detection method based on edge detection | |
| Wang et al. | Human interaction recognition based on sparse representation of feature covariance matrices | |
| CN114581485A (en) | Target tracking method based on language modeling pattern twin network | |
| Yan et al. | Crowd counting via scale-adaptive convolutional neural network in extremely dense crowd images | |
| Ge et al. | A visual tracking algorithm combining parallel network and dual attention-aware mechanism | |
| CN112989952B (en) | Crowd density estimation method and device based on mask guidance | |
| Sun et al. | YOLOv7-FIRE: A tiny-fire identification and detection method applied on UAV | |
| Jiang et al. | A Closed-Loop Detection Algorithm for Indoor Simultaneous Localization and Mapping Based on You Only Look Once v3. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |