CN108121986A - Object detection method and device, computer installation and computer readable storage medium - Google Patents
Object detection method and device, computer installation and computer readable storage medium Download PDFInfo
- Publication number
- CN108121986A CN108121986A CN201711484723.3A CN201711484723A CN108121986A CN 108121986 A CN108121986 A CN 108121986A CN 201711484723 A CN201711484723 A CN 201711484723A CN 108121986 A CN108121986 A CN 108121986A
- Authority
- CN
- China
- Prior art keywords
- region
- target
- neural networks
- network
- convolutional neural
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Image Analysis (AREA)
Abstract
A kind of object detection method, the described method includes：Training sample set is obtained, the training sample set includes multiple target images for being labeled with target location and target angle type；It is trained using the training sample set pair acceleration region convolutional neural networks model, obtains trained acceleration region convolutional neural networks model；Obtain image to be detected；Target detection is carried out to described image to be detected using the trained acceleration region convolutional neural networks model, obtains the target area of described image to be detected and the target angle type of the target area.The present invention also provides a kind of object detecting device, computer installation and readable storage medium storing program for executing.The present invention can realize the target detection of quick high detection rate.
Description
Technical field
The present invention relates to technical field of image processing, and in particular to a kind of object detection method and device, computer installation
And computer readable storage medium.
Background technology
Existing target detection technique includes the target detection of the complex characteristic based on simple pixel characteristic or hand-designed.
Using simple pixel characteristic, such as representative HAAR, pixel value difference etc., although computational efficiency is high, real-time is good,
The robustness of the factors such as the background variation for complicated variety is poor, is short of in accuracy of detection.And based on hand-designed
Complex characteristic, such as HOG in DPM etc., although feature representation is more preferable, robustness is stronger, because GPU cannot be used
Accelerate, calculated on CPU complicated, it is difficult to reach the requirement of real-time.
Existing target detection technique further includes the target detection based on convolutional neural networks.Based on convolutional neural networks
Although object detection method improves the precision of detection, but the thing followed is the significantly promotion of calculation amount.Although GPU is calculated
The computational problem of extraction convolution feature is solved, but candidate region extraction still expends for quite a long time.It is further, since entire
Scheme is first to extract candidate region, then the frame flow classified, and leads to not realize and detect end to end, applies
It is relatively cumbersome.
Further, since shooting angle is different, bigger variation can occur on the image for the appearance of target object, existing
Target detection technique does not consider the problems of shooting angle, and the recall rate for causing target is relatively low.
The content of the invention
In view of the foregoing, it is necessary to propose a kind of object detection method and device, computer installation and computer-readable
Storage medium can realize the target detection of quick high detection rate.
The first aspect of the application provides a kind of object detection method, the described method includes：
Training sample set is obtained, the training sample set includes multiple mesh for being labeled with target location and target angle type
Logo image；
It is trained using the training sample set pair acceleration region convolutional neural networks model, obtains trained acceleration
Region convolutional neural networks model, the acceleration region convolutional neural networks model include region and suggest network and fast area volume
Product neutral net, the region suggest that network and the fast area convolutional neural networks share convolutional layer, and the convolutional layer carries
The training sample is taken to concentrate the characteristic pattern of each target image, the region suggests network according to obtaining the characteristic pattern
The target angle type of candidate region and the candidate region in each target image, the fast area convolutional Neural net
Network is screened and adjusted to the candidate region according to the characteristic pattern, obtain each target image target area and
The target angle type of the target area；
Obtain image to be detected；
Target is carried out to described image to be detected using in the trained acceleration region convolutional neural networks model
Detection, obtains the target area of described image to be detected and the target angle type of the target area.
It is described to use the training sample set pair acceleration region convolutional neural networks mould in alternatively possible realization method
Type be trained including：
(1) region described in Imagenet model initializations is used to suggest network, using described in training sample set training
Suggest network in region；
(2) region in (1) after training is used to suggest that network generates the candidate region of each target image, utilizes institute
It states candidate region and trains the fast area convolutional neural networks；
(3) the fast area convolutional neural networks in (2) after training is used to initialize the region and suggest network, use institute
Stating training sample set trains the region to suggest network；
(4) region in (3) after training is used to suggest fast area convolutional neural networks described in netinit, and is kept
The convolutional layer is fixed, and the fast area convolutional neural networks are trained using the training sample set.
It is described to use the training sample set pair acceleration region convolutional neural networks mould in alternatively possible realization method
Type be trained including：
Network and the fast area convolutional neural networks, which are trained, is suggested to the region using back-propagation algorithm,
The network parameter that network and the fast area convolutional neural networks are suggested in the region is adjusted in training process, makes loss function
It minimizes, wherein the loss function includes target classification loss, angle Classification Loss and returns loss.
In alternatively possible realization method, the acceleration region convolutional neural networks model uses ZF frames, the area
Suggest that network and the fast area convolutional neural networks share 5 convolutional layers in domain.
Negative sample difficulty example is added in alternatively possible realization method, in the training of the fast area convolutional network to excavate
Method.
The second aspect of the application provides a kind of object detecting device, and described device includes：
First acquisition unit, for obtaining training sample set, the training sample set is labeled with target location including multiple
With the target image of target angle type；
Training unit for being trained using the training sample set pair acceleration region convolutional neural networks model, is obtained
To trained acceleration region convolutional neural networks model, the acceleration region convolutional neural networks model includes region and suggests net
Network and fast area convolutional neural networks, the region suggest that network and the fast area convolutional neural networks share convolution
Layer, the convolutional layer extract the characteristic pattern that the training sample concentrates each target image, and the region suggests network according to institute
The target angle type of candidate region and the candidate region that characteristic pattern is obtained in each target image is stated, it is described fast
Fast region convolutional neural networks are screened and adjusted to the candidate region according to the characteristic pattern, obtain each target
The target area of image and the target angle type of the target area；
Second acquisition unit, for obtaining image to be detected；
Detection unit, for using in trained acceleration region convolutional neural networks model to described image to be detected
Target detection is carried out, obtains the target area of described image to be detected and the target angle type of the target area.
In alternatively possible realization method, the training unit is specifically used for：
(1) region described in Imagenet model initializations is used to suggest network, using described in training sample set training
Suggest network in region；
(2) region in (1) after training is used to suggest that network generates the candidate region of each target image, utilizes institute
It states candidate region and trains the fast area convolutional neural networks；
(3) the fast area convolutional neural networks in (2) after training is used to initialize the region and suggest network, use institute
Stating training sample set trains the region to suggest network；
(4) region in (3) after training is used to suggest fast area convolutional neural networks described in netinit, and is kept
The convolutional layer is fixed, and the fast area convolutional neural networks are trained using the training sample set.
In alternatively possible realization method, the training unit is specifically used for：
Network and the fast area convolutional neural networks, which are trained, is suggested to the region using back-propagation algorithm,
The network parameter that network and the fast area convolutional neural networks are suggested in the region is adjusted in training process, makes loss function
It minimizes, wherein the loss function includes target classification loss, angle Classification Loss and returns loss.
The third aspect of the application provides a kind of computer installation, and the computer installation includes processor, the processing
Device is used to realize the object detection method when performing the computer program stored in memory.
The fourth aspect of the application provides a kind of computer readable storage medium, is stored thereon with computer program, described
The object detection method is realized when computer program is executed by processor.
The present invention obtains training sample set, and the training sample set is labeled with target location and target angle class including multiple
The target image of type；It is trained, is trained using the training sample set pair acceleration region convolutional neural networks model
Acceleration region convolutional neural networks model, the acceleration region convolutional neural networks model includes region and suggests network and quick
Region convolutional neural networks, the region suggest that network and the fast area convolutional neural networks share convolutional layer, the volume
Lamination extracts the characteristic pattern that the training sample concentrates each target image, and the region suggests that network is obtained according to the characteristic pattern
Take the target angle type of the candidate region and the candidate region in each target image, the fast area convolution
Neutral net is screened and adjusted to the candidate region according to the characteristic pattern, obtains the target of each target image
Region and the target angle type of the target area；Obtain image to be detected；Utilize the trained acceleration region convolution
Target detection is carried out to described image to be detected in neural network model, obtains the target area and institute of described image to be detected
State the target angle type of target area.
The existing target detection based on convolutional neural networks using selective search algorithm generate candidate region, take compared with
It is more, and extracted region and target detection are separated.Present invention introduce region in acceleration region convolutional neural networks model is built
Network is discussed, candidate region is extracted using depth convolutional neural networks.After network training, by sharing convolution network parameter
Method, the characteristic pattern that image is obtained by convolutional layer can be applied to extracted region and target detection simultaneously, that is, share
The result of calculation of convolutional network so as to the speed of significantly lifting region extraction, accelerates the speed of entire testing process, realizes end
To the detection scheme at end.Also, the present invention considers the problem of shooting angle difference causes recall rate to reduce, using being labeled with mesh
The target image of mark angular type is trained acceleration region convolutional neural networks model, improves the recall rate of target.Cause
This, the present invention can realize the target detection of quick high detection rate.
Description of the drawings
Fig. 1 is the flow chart for the object detection method that the embodiment of the present invention one provides.
Fig. 2 is the schematic diagram that network is suggested in region.
Fig. 3 is the structure chart of object detecting device provided by Embodiment 2 of the present invention.
Fig. 4 is the schematic diagram for the computer installation that the embodiment of the present invention three provides.
Specific embodiment
It is to better understand the objects, features and advantages of the present invention, below in conjunction with the accompanying drawings and specific real
Applying example, the present invention will be described in detail.It should be noted that in the case where there is no conflict, embodiments herein and embodiment
In feature can be mutually combined.
Elaborate many details in the following description to facilitate a thorough understanding of the present invention, described embodiment only
Only it is part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, ordinary skill
Personnel's all other embodiments obtained without making creative work, belong to the scope of protection of the invention.
Unless otherwise defined, all of technologies and scientific terms used here by the article is with belonging to technical field of the invention
The normally understood meaning of technical staff is identical.Term used in the description of the invention herein is intended merely to description tool
The purpose of the embodiment of body, it is not intended that in the limitation present invention.
Preferably, object detection method of the invention is applied in one or more computer installation.The computer
Device be it is a kind of can be according to the instruction for being previously set or storing, the automatic equipment for carrying out numerical computations and/or information processing,
Hardware includes but not limited to microprocessor, application-specific integrated circuit (Application Specific Integrated
Circuit, ASIC), programmable gate array (Field-Programmable Gate Array, FPGA), digital processing unit
(Digital Signal Processor, DSP), embedded device etc..
The computer installation can be that the calculating such as desktop PC, notebook, palm PC and cloud server are set
It is standby.The computer installation can with user by modes such as keyboard, mouse, remote controler, touch tablet or voice-operated devices into pedestrian
Machine interacts.
Embodiment one
Fig. 1 is the flow chart for the object detection method that the embodiment of the present invention one provides.The object detection method is applied to
Computer installation.The object detection method can detect the position of goal-selling in image (such as vehicle, ship), and can
To detect the angular type of goal-selling in image (such as front, side, back side).
As shown in Figure 1, the object detection method specifically includes following steps：
101：Obtain training sample set.
The training sample set includes multiple target images for being labeled with target location and target angle type.The target
Image is the image for including goal-selling (such as ship, vehicle etc.).The target image can include one or more default
Target.The target location represents the position of goal-selling in the target image.The target angle type represents goal-selling
Shooting angle (such as front, the back side, side).
In one embodiment, the training sample set includes about 10000 target images.The target location can be with
It is labeled as [x, y, w, h], x, y represents the top left co-ordinate of target area, and w represents the width of target area, and h represents target area
Height.The target angle type includes positive angular type, flank angle type and back angle type.It is for example, described
When object detection method is used to be detected ship, if target image is the direct picture of ship, the target angle marked
Type is positive angular type；If target image is the side image of ship, the target angle type marked is flank angle
Type；If target image is the back side image of ship, the target angle type marked is back angle type.
102：Use training sample set pair acceleration region convolutional neural networks model (the Faster Region-based
Convolution Neural Network, Faster R-CNN) it is trained, obtain trained acceleration region convolutional Neural
Network model.
The acceleration region convolutional neural networks model include region suggest network (Region Proposal Network,
) and fast area convolutional neural networks (Fast Region-based Convolution Neural Network, Fast RPN
R-CNN).It needs to suggest region network and fast convolution network carry out alternately training.
Network is suggested in the region and the fast area convolutional neural networks have shared convolutional layer, and the convolutional layer is used
In the characteristic pattern of extraction image.Suggest that network generates candidate region and the candidate regions of image according to the characteristic pattern in the region
The target angle type in domain, and the target angle type of the candidate region of generation and candidate region is inputted into the fast area
Convolutional neural networks.The fast area convolutional neural networks are screened and adjusted to the candidate region according to the characteristic pattern
It is whole, obtain the target area of image and the target angle type of target area.
Specifically, in training, the convolutional layer extracts the characteristic pattern that the training sample concentrates each target image, institute
It states region and suggests that network obtains candidate region and the candidate region in each target image according to the characteristic pattern
Target angle type, the fast area convolutional neural networks according to the characteristic pattern to the candidate region carry out screening and
Adjustment, obtains the target area of each target image and the target angle type of the target area.
In a preferred embodiment, the acceleration region convolutional neural networks model uses ZF frames, and the region is suggested
Network and the fast area convolutional neural networks share 5 convolutional layers.
In one embodiment, the target image that training sample is concentrated can be the image of arbitrary dimension, in input institute
Target image is scaled to the image of unified size (such as 1000*600) before stating convolutional layer.In one embodiment, institute
The length and width for stating the characteristic pattern of convolutional layer extraction reduce 16 times compared with input picture, and the depth of characteristic pattern is 256.
In one embodiment, being trained using training sample set pair acceleration region convolutional neural networks model can be with
Including：
(1) region described in Imagenet model initializations is used to suggest network, using described in training sample set training
Suggest network in region.
(2) region in (1) after training is used to suggest that network generation training sample concentrates the candidate regions of each target image
The fast area convolutional neural networks are trained in domain using the candidate region.At this point, network and fast area volume are suggested in region
Product neutral net shares convolutional layer not yet.
(3) the fast area convolutional neural networks in (2) after training is used to initialize the region and suggest network, use instruction
Practicing sample set trains the region to suggest network.
(4) region in (3) after training is used to suggest fast area convolutional neural networks described in netinit, and is kept
The convolutional layer is fixed, and the fast area convolutional neural networks are trained using training sample set.At this point, region suggest network and
Fast area convolutional neural networks share identical convolutional layer, constitute a unified network.
Fig. 2 is the schematic diagram that network is suggested in region.
After image is by shared convolutional layer, the characteristic pattern of image is obtained.With default size (such as 3*3) on characteristic pattern
Sliding window slided according to default step-length (such as step-length be 1), the every position of sliding window corresponds to a central point.When
When sliding window slides into a position, to the default scale (such as 3 kinds of scales 128,256,512) of the central point of position application and
Default length-width ratio (such as 3 kinds of length-width ratios 1:1、1:2、2:1) anchor frame obtains default quantity (such as 9) candidate region.Pass through
Each sliding window is mapped to the feature vector of a low-dimensional by one convolutional layer (convolutional layer is cascaded with shared convolutional layer)
In (such as feature vector of 256-d or 512-d).This feature vector is exported to three full articulamentums at the same level, one is mesh
Mark classification layer, one is angle classification layer, and one is that border returns layer.The target classification of target classification layer output candidate region obtains
Point, it is target (i.e. prospect) or background to be used to indicate candidate region.Candidate region belongs to prospect or background, depending on candidate
The registration in region and the target area (region determined by the target location marked) of mark, registration are more than some threshold value
Then it is positioned as prospect, registration is then positioned as background less than threshold value.The angle classification score of angle classification layer output candidate region,
It is used to indicate the target angle type of candidate region.Border returns the position of the candidate region after layer output fine tuning, for waiting
The border of favored area is finely adjusted.
Region suggests that the candidate region that network is chosen is more, if can have been screened according to the target classification score of candidate region
The candidate region of dry highest scoring is input to fast area convolutional neural networks, to accelerate the speed of training and detection.
In order to train region suggest network, give each candidate region distribute a label, the label include positive label and
Negative label, positive label can distribute to two class candidate regions：(1) with some real goal (Ground Truth, GT) bounding box
There is the candidate region that highest IoU (the ratio between Intersection over Union, intersection union) is overlapped；(2) with arbitrary GT sides
Boundary's frame has the candidate region of the IoU overlappings more than 0.7.For a GT bounding box, positive label may be distributed to multiple candidate regions
Domain.Negative label distribute to be below with the IoU ratios of all GT bounding boxes 0.3 candidate region.Non- just non-negative candidate region
There is no any effect to training objective.
Region suggests that the training of network is trained using back-propagation algorithm, and adjustment region suggests network in training process
Network parameter, minimize loss function.Suggest the prediction confidence of the candidate region of neural network forecast in loss function indicating area
Degree and the difference of true confidence level.In the present embodiment, loss function includes target classification loss, angle Classification Loss and recurrence
Lose three parts.
The loss function of image can be defined as：
Wherein, i is the index of candidate region in a training batch (mini-batch).
It is the target classification loss of candidate region.NclsFor the size of training batch, such as 256.piIt is i-th
A candidate region is the prediction probability of target.It is GT labels, if candidate region is just (label distributed is positive label, is claimed
For positive candidate region),For 1；If candidate region is negative (label distributed is negative label, referred to as negative candidate region),For
0。It may be calculated
It is the angle Classification Loss of candidate region,Meaning be referred to
It is the recurrence loss of candidate region.λ is balance weight, can be taken as 10.NregFor candidate region
Quantity.It may be calculatedtiA coordinate vector, i.e. ti=(tx,ty,tw,
th), represent that 4 of candidate region parameterize coordinates (such as the coordinate in the candidate region upper left corner and width, height).Be with
The coordinate vector of the corresponding GT bounding boxes in positive candidate region, i.e.,(such as real goal region upper left
The coordinate and width at angle, height).R is the loss function (smooth with robustnessL1), it is defined as：
Above-described embodiment considers the problem of shooting angle difference causes recall rate to reduce, in acceleration region convolutional Neural net
Using the loss function for including angle Classification Loss in the training of network model, candidate regions are calculated according to the target angle type of prediction
The angle Classification Loss in domain improves the recall rate of target.
The above-mentioned training method for suggesting network for region.The training method of fast area convolutional network is referred to region and builds
The training method of network is discussed, details are not described herein again.
In the present embodiment, negative sample difficulty example is added in the training of fast area convolutional network and excavates (Hard
Negative Mining, HNM) method.For being wrongly classified as the negative sample of positive sample by fast area convolutional network (i.e.
Difficult example), the information record of these negative samples is got off, during next repetitive exercise, these negative samples are inputted again
It is concentrated to training sample, and increases the weight of its loss, enhanced its influence to grader, can so ensure ceaselessly pin
Classify to the negative sample being more difficult to so that from the easier to the more advanced, the sample distribution covered is also more various for the feature that grader is acquired
Property.
103：Obtain image to be detected.
Image to be detected is to include the image of goal-selling (such as ship).Goal-selling is the detection in image to be detected
Object.For example, when carrying out ship detection to image to be detected, goal-selling is the ship in image to be detected.
Image to be detected can be the image received from external equipment, such as the ship figure of the camera shooting near harbour
Picture receives the ship image from the camera.
Alternatively, image to be detected can be the image of the computer installation shooting, such as computer installation shooting
Ship image.
Alternatively, image to be detected can also be the image read from the memory of the computer installation, such as from institute
State the ship image read in the memory of computer installation.
104：Image to be detected is detected using trained acceleration region convolutional neural networks model, is obtained to be checked
The target area of altimetric image and the target angle type of the target area.
Specifically, the region suggests that the convolutional layer extraction that network and the fast area convolutional neural networks are shared is to be checked
The characteristic pattern of altimetric image.Suggest that network obtains candidate region and institute in image to be detected according to the characteristic pattern in the region
State the target angle type of candidate region.The fast area convolutional neural networks are according to the characteristic pattern to the candidate region
It is screened and is adjusted, obtain the target area of image to be detected and the target angle type of the target area.
The object detection method of embodiment one obtains training sample set, and the training sample set is labeled with target including multiple
Position and the target image of target angle type；It is carried out using the training sample set pair acceleration region convolutional neural networks model
Training, obtains trained acceleration region convolutional neural networks model, and the acceleration region convolutional neural networks model includes area
Network and fast area convolutional neural networks are suggested in domain, and the region suggests that network and the fast area convolutional neural networks are total to
Convolutional layer is enjoyed, the convolutional layer extracts the characteristic pattern that the training sample concentrates each target image, and network is suggested in the region
The target angle type of the candidate region and the candidate region in each target image is obtained according to the characteristic pattern,
The fast area convolutional neural networks are screened and adjusted to the candidate region according to the characteristic pattern, are obtained described each
The target area of a target image and the target angle type of the target area；Obtain image to be detected；Utilize the training
Target detection is carried out to described image to be detected in good acceleration region convolutional neural networks model, obtains the mapping to be checked
The target area of picture and the target angle type of the target area.
The existing target detection based on convolutional neural networks using selective search algorithm generate candidate region, take compared with
It is more, and extracted region and target detection are separated.The object detection method of embodiment one is in acceleration region convolutional neural networks
Introduce region suggests network in model, and candidate region is extracted using depth convolutional neural networks.After network training, pass through
The method of shared convolution network parameter, the characteristic pattern that image is obtained by convolutional layer can be applied to extracted region and target simultaneously
Detection, that is, the result of calculation of shared convolutional network, so as to the speed of significantly lifting region extraction, accelerate entire detection stream
The speed of journey realizes detection scheme end to end.Also, the object detection method of embodiment one considers shooting angle difference and draws
The problem of recall rate reduces is played, using being labeled with the target image of target angle type to acceleration region convolutional neural networks model
It is trained, improves the recall rate of target.Therefore, the object detection method of embodiment one can realize quick high detection rate
Target detection.
Embodiment two
Fig. 3 is the structure chart of object detecting device provided by Embodiment 2 of the present invention.As shown in figure 3, the target detection
Device 10 can include：First acquisition unit 301, training unit 302, second acquisition unit 303, detection unit 304.
First acquisition unit 301, for obtaining training sample set.
The training sample set includes multiple target images for being labeled with target location and target angle type.The target
Image is the image for including goal-selling (such as ship, vehicle etc.).The target image can include one or more default
Target.The target location represents the position of goal-selling in the target image.The target angle type represents goal-selling
Shooting angle (such as front, the back side, side).
In one embodiment, the training sample set includes about 10000 target images.The target location can be with
It is labeled as [x, y, w, h], x, y represents the top left co-ordinate of target area, and w represents the width of target area, and h represents target area
Height.The target angle type includes positive angular type, flank angle type and back angle type.It is for example, described
When object detection method is used to be detected ship, if target image is the direct picture of ship, the target angle marked
Type is positive angular type；If target image is the side image of ship, the target angle type marked is flank angle
Type；If target image is the back side image of ship, the target angle type marked is back angle type.
Training unit 302, for using the training sample set pair acceleration region convolutional neural networks model (Faster
Region-based Convolution Neural Network, Faster R-CNN) it is trained, obtain trained add
Fast region convolutional neural networks model.
The acceleration region convolutional neural networks model include region suggest network (Region Proposal Network,
) and fast area convolutional neural networks (Fast Region-based Convolution Neural Network, Fast RPN
R-CNN).It needs to suggest region network and fast convolution network carry out alternately training.
Network is suggested in the region and the fast area convolutional neural networks have shared convolutional layer, and the convolutional layer is used
In the characteristic pattern of extraction image.Suggest that network generates candidate region and the candidate regions of image according to the characteristic pattern in the region
The target angle type in domain, and the target angle type of the candidate region of generation and candidate region is inputted into the fast area
Convolutional neural networks.The fast area convolutional neural networks are screened and adjusted to the candidate region according to the characteristic pattern
It is whole, obtain the target area of image and the target angle type of target area.
Specifically, in training, the convolutional layer extracts the characteristic pattern that the training sample concentrates each target image, institute
It states region and suggests that network obtains candidate region and the candidate region in each target image according to the characteristic pattern
Target angle type, the fast area convolutional neural networks according to the characteristic pattern to the candidate region carry out screening and
Adjustment, obtains the target area of each target image and the target angle type of the target area.
In a preferred embodiment, the acceleration region convolutional neural networks model uses ZF frames, and the region is suggested
Network and the fast area convolutional neural networks share 5 convolutional layers.
In one embodiment, the target image that training sample is concentrated can be the image of arbitrary dimension, in input institute
Target image is scaled to the image of unified size (such as 1000*600) before stating convolutional layer.In one embodiment, institute
The length and width for stating the characteristic pattern of convolutional layer extraction reduce 16 times compared with input picture, and the depth of characteristic pattern is 256.
In one embodiment, being trained using training sample set pair acceleration region convolutional neural networks model can be with
Including：
(1) region described in Imagenet model initializations is used to suggest network, using described in training sample set training
Suggest network in region.
(2) region in (1) after training is used to suggest that network generation training sample concentrates the candidate regions of each target image
The fast area convolutional neural networks are trained in domain using the candidate region.At this point, network and fast area volume are suggested in region
Product neutral net shares convolutional layer not yet.
(3) the fast area convolutional neural networks in (2) after training is used to initialize the region and suggest network, use instruction
Practicing sample set trains the region to suggest network.
(4) region in (3) after training is used to suggest fast area convolutional neural networks described in netinit, and is kept
The convolutional layer is fixed, and the fast area convolutional neural networks are trained using training sample set.At this point, region suggest network and
Fast area convolutional neural networks share identical convolutional layer, constitute a unified network.
Fig. 2 is the schematic diagram that network is suggested in region.
After image is by shared convolutional layer, the characteristic pattern of image is obtained.With default size (such as 3*3) on characteristic pattern
Sliding window slided according to default step-length (such as step-length be 1), the every position of sliding window corresponds to a central point.When
When sliding window slides into a position, to the default scale (such as 3 kinds of scales 128,256,512) of the central point of position application and
Default length-width ratio (such as 3 kinds of length-width ratios 1:1、1:2、2:1) anchor frame obtains default quantity (such as 9) candidate region.Pass through
Each sliding window is mapped to the feature vector of a low-dimensional by one convolutional layer (convolutional layer is cascaded with shared convolutional layer)
In (such as feature vector of 256-d or 512-d).This feature vector is exported to three full articulamentums at the same level, one is mesh
Mark classification layer, one is angle classification layer, and one is that border returns layer.The target classification of target classification layer output candidate region obtains
Point, it is target (i.e. prospect) or background to be used to indicate candidate region.Candidate region belongs to prospect or background, depending on candidate
The registration in region and the target area (region determined by the target location marked) of mark, registration are more than some threshold value
Then it is positioned as prospect, registration is then positioned as background less than threshold value.The angle classification score of angle classification layer output candidate region,
It is used to indicate the target angle type of candidate region.Border returns the position of the candidate region after layer output fine tuning, for waiting
The border of favored area is finely adjusted.
Region suggests that the candidate region that network is chosen is more, if can have been screened according to the target classification score of candidate region
The candidate region of dry highest scoring is input to fast area convolutional neural networks, to accelerate the speed of training and detection.
In order to train region suggest network, give each candidate region distribute a label, the label include positive label and
Negative label, positive label can distribute to two class candidate regions：(1) with some real goal (Ground Truth, GT) bounding box
There is the candidate region that highest IoU (the ratio between Intersection over Union, intersection union) is overlapped；(2) with arbitrary GT sides
Boundary's frame has the candidate region of the IoU overlappings more than 0.7.For a GT bounding box, positive label may be distributed to multiple candidate regions
Domain.Negative label distribute to be below with the IoU ratios of all GT bounding boxes 0.3 candidate region.Non- just non-negative candidate region
There is no any effect to training objective.
Region suggests that the training of network is trained using back-propagation algorithm, and adjustment region suggests network in training process
Network parameter, minimize loss function.Suggest the prediction confidence of the candidate region of neural network forecast in loss function indicating area
Degree and the difference of true confidence level.In the present embodiment, loss function includes target classification loss, angle Classification Loss and recurrence
Lose three parts.
The loss function of image can be defined as：
Wherein, i is the index of candidate region in a training batch (mini-batch).
It is the target classification loss of candidate region.NclsFor the size of training batch, such as 256.piIt is i-th
A candidate region is the prediction probability of target.It is GT labels, if candidate region is just (label distributed is positive label, is claimed
For positive candidate region),For 1；If candidate region is negative (label distributed is negative label, referred to as negative candidate region),For
0。It may be calculated
It is the angle Classification Loss of candidate region,Meaning be referred to
It is the recurrence loss of candidate region.λ is balance weight, can be taken as 10.NregFor candidate region
Quantity.It may be calculatedtiA coordinate vector, i.e. ti=(tx,ty,tw,
th), represent that 4 of candidate region parameterize coordinates (such as the coordinate in the candidate region upper left corner and width, height).Be with
The coordinate vector of the corresponding GT bounding boxes in positive candidate region, i.e.,(such as real goal region upper left
The coordinate and width at angle, height).R is the loss function (smooth with robustnessL1), it is defined as：
Above-described embodiment considers the problem of shooting angle difference causes recall rate to reduce, in acceleration region convolutional Neural net
Using the loss function for including angle Classification Loss in the training of network model, candidate regions are calculated according to the target angle type of prediction
The angle Classification Loss in domain improves the recall rate of target.
The above-mentioned training method for suggesting network for region.The training method of fast area convolutional network is referred to region and builds
The training method of network is discussed, details are not described herein again.
In the present embodiment, negative sample difficulty example is added in the training of fast area convolutional network and excavates (Hard
Negative Mining, HNM) method.For being wrongly classified as the negative sample of positive sample by fast area convolutional network (i.e.
Difficult example), the information record of these negative samples is got off, during next repetitive exercise, these negative samples are inputted again
It is concentrated to training sample, and increases the weight of its loss, enhanced its influence to grader, can so ensure ceaselessly pin
Classify to the negative sample being more difficult to so that from the easier to the more advanced, the sample distribution covered is also more various for the feature that grader is acquired
Property.
Second acquisition unit 303, for obtaining image to be detected.
Image to be detected is to include the image of goal-selling (such as ship).Goal-selling is the detection in image to be detected
Object.For example, when carrying out ship detection to image to be detected, goal-selling is the ship in image to be detected.
Image to be detected can be the image received from external equipment, such as the ship figure of the camera shooting near harbour
Picture receives the ship image from the camera.
Alternatively, image to be detected can be the image of the computer installation shooting, such as computer installation shooting
Ship image.
Alternatively, image to be detected can also be the image read from the memory of the computer installation, such as from institute
State the ship image read in the memory of computer installation.
Detection unit 304, for being carried out using trained acceleration region convolutional neural networks model to image to be detected
Detection, obtains the target area of image to be detected and the target angle type of the target area.
Specifically, the region suggests that the convolutional layer extraction that network and the fast area convolutional neural networks are shared is to be checked
The characteristic pattern of altimetric image.Suggest that network obtains candidate region and institute in image to be detected according to the characteristic pattern in the region
State the target angle type of candidate region.The fast area convolutional neural networks are according to the characteristic pattern to the candidate region
It is screened and is adjusted, obtain the target area of image to be detected and the target angle type of the target area.
Embodiment two obtains training sample set, and the training sample set is labeled with target location and target angle including multiple
The target image of type；It is trained, is trained using the training sample set pair acceleration region convolutional neural networks model
Good acceleration region convolutional neural networks model, the acceleration region convolutional neural networks model is including region suggestion network and soon
Fast region convolutional neural networks, the region suggests that network and the fast area convolutional neural networks share convolutional layer, described
Convolutional layer extracts the characteristic pattern that the training sample concentrates each target image, and the region suggests network according to the characteristic pattern
Obtain the target angle type of the candidate region and the candidate region in each target image, the fast area volume
Product neutral net is screened and adjusted to the candidate region according to the characteristic pattern, obtains the mesh of each target image
Mark the target angle type of region and the target area；Obtain image to be detected；It is rolled up using the trained acceleration region
Product neural network model in described image to be detected carry out target detection, obtain described image to be detected target area and
The target angle type of the target area.
The existing target detection based on convolutional neural networks using selective search algorithm generate candidate region, take compared with
It is more, and extracted region and target detection are separated.The introduce region in acceleration region convolutional neural networks model of embodiment two
It is recommended that network, candidate region is extracted using depth convolutional neural networks.After network training, joined by shared convolutional network
Several methods, the characteristic pattern that image is obtained by convolutional layer can be applied to extracted region and target detection simultaneously, that is, common
The result of calculation of convolutional network is enjoyed, so as to the speed of significantly lifting region extraction, accelerates the speed of entire testing process, realizes
Detection scheme end to end.Also, embodiment two considers the problem of shooting angle difference causes recall rate to reduce, and uses mark
The target image for having target angle type is trained acceleration region convolutional neural networks model, improves the detection of target
Rate.Therefore, embodiment two can realize the target detection of quick high detection rate.
Embodiment three
Fig. 4 is the schematic diagram for the computer installation that the embodiment of the present invention three provides.The computer installation 1 includes memory
20th, processor 30 and the computer program 40 that can be run in the memory 20 and on the processor 30, example are stored in
Such as object detection program.The processor 30 is realized when performing the computer program 40 in above-mentioned object detection method embodiment
The step of, such as step 101~104 shown in FIG. 1.Alternatively, the processor 30 is realized when performing the computer program 40
The function of each module/unit in above device embodiment, such as the unit 301~304 in Fig. 3.
Illustratively, the computer program 40 can be divided into one or more module/units, it is one or
Multiple module/units are stored in the memory 20, and are performed by the processor 30, to complete the present invention.Described one
A or multiple module/units can be the series of computation machine program instruction section that can complete specific function, which is used for
Implementation procedure of the computer program 40 in the computer installation 1 is described.For example, the computer program 40 can be by
First acquisition unit 301, training unit 302, second acquisition unit 303, the detection unit 304 being divided into Fig. 3, each unit tool
Body function is referring to embodiment two.
The computer installation 1 can be that the calculating such as desktop PC, notebook, palm PC and cloud server are set
It is standby.It will be understood by those skilled in the art that the schematic diagram 4 is only the example of computer installation 1, do not form to computer
The restriction of device 1 can include either combining some components or different components, example than illustrating more or fewer components
Such as described computer installation 1 can also include input-output equipment, network access equipment, bus.
Alleged processor 30 can be central processing unit (Central Processing Unit, CPU), can also be
Other general processors, digital signal processor (Digital Signal Processor, DSP), application-specific integrated circuit
(Application Specific Integrated Circuit, ASIC), ready-made programmable gate array (Field-
Programmable Gate Array, FPGA) either other programmable logic device, discrete gate or transistor logic,
Discrete hardware components etc..General processor can be microprocessor or the processor 30 can also be any conventional processor
Deng the processor 30 is the control centre of the computer installation 1, utilizes various interfaces and connection entire computer dress
Put 1 various pieces.
The memory 20 can be used for storing the computer program 40 and/or module/unit, and the processor 30 passes through
The computer program and/or module/unit and calling that operation or execution are stored in the memory 20 are stored in memory
Data in 20 realize the various functions of the computer installation 1.The memory 20 can mainly include storing program area and deposit
Store up data field, wherein, storing program area can storage program area, the application program needed at least one function (for example broadcast by sound
Playing function, image player function etc.) etc.；Storage data field can be stored uses created data (ratio according to computer installation 1
Such as voice data, phone directory) etc..In addition, memory 20 can include high-speed random access memory, can also include non-easy
The property lost memory, such as hard disk, memory, plug-in type hard disk, intelligent memory card (Smart Media Card, SMC), secure digital
(Secure Digital, SD) block, flash card (Flash Card), at least one disk memory, flush memory device or other
Volatile solid-state part.
If the integrated module/unit of the computer installation 1 is realized in the form of SFU software functional unit and as independently
Production marketing or in use, can be stored in a computer read/write memory medium.Based on such understanding, the present invention
It realizes all or part of flow in above-described embodiment method, relevant hardware can also be instructed by computer program come complete
Into the computer program can be stored in a computer readable storage medium, which is being executed by processor
When, it can be achieved that the step of above-mentioned each embodiment of the method.Wherein, the computer program includes computer program code, described
Computer program code can be source code form, object identification code form, executable file or some intermediate forms etc..The meter
Calculation machine readable medium can include：Can carry the computer program code any entity or device, recording medium, USB flash disk,
Mobile hard disk, magnetic disc, CD, computer storage, read-only memory (ROM, Read-Only Memory), random access memory
Device (RAM, Random Access Memory), electric carrier signal, telecommunication signal and software distribution medium etc..Need what is illustrated
It is that the content that the computer-readable medium includes can be fitted according to legislation in jurisdiction and the requirement of patent practice
When increase and decrease, such as in some jurisdictions, according to legislation and patent practice, computer-readable medium, which does not include electric carrier wave, to be believed
Number and telecommunication signal.
In several embodiments provided by the present invention, it should be understood that disclosed computer installation and method, it can be with
It realizes by another way.For example, computer installation embodiment described above is only schematical, for example, described
The division of unit is only a kind of division of logic function, can there is other dividing mode in actual implementation.
In addition, each functional unit in each embodiment of the present invention can be integrated in same treatment unit, it can also
That unit is individually physically present, can also two or more units be integrated in same unit.Above-mentioned integrated list
The form that hardware had both may be employed in member is realized, can also be realized in the form of hardware adds software function module.
It is obvious to a person skilled in the art that the invention is not restricted to the details of above-mentioned exemplary embodiment, Er Qie
In the case of without departing substantially from spirit or essential attributes of the invention, the present invention can be realized in other specific forms.Therefore, no matter
From the point of view of which point, the present embodiments are to be considered as illustrative and not restrictive, and the scope of the present invention is by appended power
Profit requirement rather than above description limit, it is intended that all by what is fallen within the meaning and scope of the equivalent requirements of the claims
Variation includes within the present invention.Any reference numeral in claim should not be considered as to the involved claim of limitation.This
Outside, it is clear that one word of " comprising " is not excluded for other units or step, and odd number is not excluded for plural number.It is stated in computer installation claim
Multiple units or computer installation can also be realized by same unit or computer installation by software or hardware.The
One, the second grade words are used to indicate names, and are not represented any particular order.
Finally it should be noted that the above embodiments are merely illustrative of the technical solutions of the present invention and it is unrestricted, although reference
The present invention is described in detail in preferred embodiment, it will be understood by those of ordinary skill in the art that, it can be to the present invention's
Technical solution is modified or equivalent substitution, without departing from the spirit and scope of technical solution of the present invention.
Claims (10)
1. a kind of object detection method, which is characterized in that the described method includes：
Training sample set is obtained, the training sample set includes multiple target figures for being labeled with target location and target angle type
Picture；
It is trained using the training sample set pair acceleration region convolutional neural networks model, obtains trained acceleration region
Convolutional neural networks model, the acceleration region convolutional neural networks model include region and suggest network and fast area convolution god
Through network, network is suggested in the region and the fast area convolutional neural networks share convolutional layer, and the convolutional layer extracts institute
The characteristic pattern that training sample concentrates each target image is stated, it is described each that the region suggests that network is obtained according to the characteristic pattern
The target angle type of candidate region and the candidate region in target image, the fast area convolutional neural networks root
The candidate region is screened and adjusted according to the characteristic pattern, obtains the target area of each target image and described
The target angle type of target area；
Obtain image to be detected；
Target detection is carried out to described image to be detected using in the trained acceleration region convolutional neural networks model,
Obtain the target area of described image to be detected and the target angle type of the target area.
2. the method as described in claim 1, which is characterized in that described to use training sample set pair acceleration region convolution god
Through network model be trained including：
(1) region described in Imagenet model initializations is used to suggest network, the region is trained using the training sample set
It is recommended that network；
(2) region in (1) after training is used to suggest that network generates the candidate region of each target image, utilizes the time
Favored area trains the fast area convolutional neural networks；
(3) the fast area convolutional neural networks in (2) after training is used to initialize the region and suggest network, use the instruction
Practicing sample set trains the region to suggest network；
(4) region in (3) after training is used to suggest fast area convolutional neural networks described in netinit, and described in holding
Convolutional layer is fixed, and the fast area convolutional neural networks are trained using the training sample set.
3. the method as described in claim 1, which is characterized in that described to use training sample set pair acceleration region convolution god
Through network model be trained including：
Network and the fast area convolutional neural networks, which are trained, is suggested to the region using back-propagation algorithm, training
The network parameter that network and the fast area convolutional neural networks are suggested in the region is adjusted in the process, makes loss function minimum
Change, wherein the loss function includes target classification loss, angle Classification Loss and returns loss.
4. method as claimed any one in claims 1 to 3, which is characterized in that the acceleration region convolutional neural networks mould
Type uses ZF frames, and the region suggests that network and the fast area convolutional neural networks share 5 convolutional layers.
5. method as claimed any one in claims 1 to 3, which is characterized in that the training of the fast area convolutional network
Middle addition negative sample difficulty example method for digging.
6. a kind of object detecting device, which is characterized in that described device includes：
First acquisition unit, for obtaining training sample set, the training sample set is labeled with target location and mesh including multiple
Mark the target image of angular type；
Training unit for being trained using the training sample set pair acceleration region convolutional neural networks model, is instructed
The acceleration region convolutional neural networks model perfected, the acceleration region convolutional neural networks model include region suggest network and
Fast area convolutional neural networks, the region suggest that network and the fast area convolutional neural networks share convolutional layer, institute
It states convolutional layer and extracts the characteristic pattern that the training sample concentrates each target image, the region suggests network according to the feature
Figure obtains the target angle type of the candidate region and the candidate region in each target image, the fast area
Convolutional neural networks are screened and adjusted to the candidate region according to the characteristic pattern, obtain each target image
Target area and the target angle type of the target area；
Second acquisition unit, for obtaining image to be detected；
Detection unit, for being carried out using in trained acceleration region convolutional neural networks model to described image to be detected
Target detection obtains the target area of described image to be detected and the target angle type of the target area.
7. device as claimed in claim 6, which is characterized in that the training unit is specifically used for：
(1) region described in Imagenet model initializations is used to suggest network, the region is trained using the training sample set
It is recommended that network；
(2) region in (1) after training is used to suggest that network generates the candidate region of each target image, utilizes the time
Favored area trains the fast area convolutional neural networks；
(3) the fast area convolutional neural networks in (2) after training is used to initialize the region and suggest network, use the instruction
Practicing sample set trains the region to suggest network；
(4) region in (3) after training is used to suggest fast area convolutional neural networks described in netinit, and described in holding
Convolutional layer is fixed, and the fast area convolutional neural networks are trained using the training sample set.
8. device as claimed in claim 6, which is characterized in that the training unit is specifically used for：
Network and the fast area convolutional neural networks, which are trained, is suggested to the region using back-propagation algorithm, training
The network parameter that network and the fast area convolutional neural networks are suggested in the region is adjusted in the process, makes loss function minimum
Change, wherein the loss function includes target classification loss, angle Classification Loss and returns loss.
9. a kind of computer installation, it is characterised in that：The computer installation includes processor, and the processor is deposited for performing
The object detection method as any one of claim 1-5 is realized during the computer program stored in reservoir.
10. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that：The computer program
The object detection method as any one of claim 1-5 is realized when being executed by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711484723.3A CN108121986B (en) | 2017-12-29 | 2017-12-29 | Object detection method and device, computer device and computer readable storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201711484723.3A CN108121986B (en) | 2017-12-29 | 2017-12-29 | Object detection method and device, computer device and computer readable storage medium |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN108121986A true CN108121986A (en) | 2018-06-05 |
| CN108121986B CN108121986B (en) | 2019-12-17 |
Family
ID=62230688
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201711484723.3A Active CN108121986B (en) | 2017-12-29 | 2017-12-29 | Object detection method and device, computer device and computer readable storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN108121986B (en) |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108921099A (en) * | 2018-07-03 | 2018-11-30 | 常州大学 | Moving ship object detection method in a kind of navigation channel based on deep learning |
| CN109063740A (en) * | 2018-07-05 | 2018-12-21 | 高镜尧 | The detection model of ultrasonic image common-denominator target constructs and detection method, device |
| CN109101966A (en) * | 2018-06-08 | 2018-12-28 | 中国科学院宁波材料技术与工程研究所 | Workpiece identification positioning and posture estimation system and method based on deep learning |
| CN109359683A (en) * | 2018-10-15 | 2019-02-19 | 百度在线网络技术（北京）有限公司 | Object detection method, device, terminal and computer readable storage medium |
| CN109376619A (en) * | 2018-09-30 | 2019-02-22 | 中国人民解放军陆军军医大学 | A kind of cell detection method |
| CN109583445A (en) * | 2018-11-26 | 2019-04-05 | 平安科技（深圳）有限公司 | Character image correction processing method, device, equipment and storage medium |
| CN109583396A (en) * | 2018-12-05 | 2019-04-05 | 广东亿迅科技有限公司 | A kind of region prevention method, system and terminal based on CNN two stages human testing |
| CN109784385A (en) * | 2018-12-29 | 2019-05-21 | 广州海昇计算机科技有限公司 | A kind of commodity automatic identifying method, system, device and storage medium |
| CN109886998A (en) * | 2019-01-23 | 2019-06-14 | 平安科技（深圳）有限公司 | Multi-object tracking method, device, computer installation and computer storage medium |
| CN109934088A (en) * | 2019-01-10 | 2019-06-25 | 海南大学 | Sea ship discrimination method based on deep learning |
| CN110288082A (en) * | 2019-06-05 | 2019-09-27 | 北京字节跳动网络技术有限公司 | Convolutional neural networks model training method, device and computer readable storage medium |
| CN110428357A (en) * | 2019-08-09 | 2019-11-08 | 厦门美图之家科技有限公司 | The detection method of watermark, device, electronic equipment and storage medium in image |
| CN110443244A (en) * | 2019-08-12 | 2019-11-12 | 深圳市捷顺科技实业股份有限公司 | A kind of method and relevant apparatus of graphics process |
| CN110599456A (en) * | 2019-08-13 | 2019-12-20 | 杭州智团信息技术有限公司 | Method for extracting specific region of medical image |
| CN110619255A (en) * | 2018-06-19 | 2019-12-27 | 杭州海康威视数字技术股份有限公司 | Target detection method and device |
| CN110717905A (en) * | 2019-09-30 | 2020-01-21 | 上海联影智能医疗科技有限公司 | Brain image detection method, computer device, and storage medium |
| CN110796154A (en) * | 2018-08-03 | 2020-02-14 | 华为技术有限公司 | Method, device and equipment for training object detection model |
| CN110807431A (en) * | 2019-11-06 | 2020-02-18 | 上海眼控科技股份有限公司 | Object positioning method and device, electronic equipment and storage medium |
| CN110837760A (en) * | 2018-08-17 | 2020-02-25 | 北京四维图新科技股份有限公司 | Target detection method, training method and device for target detection |
| CN110991531A (en) * | 2019-12-02 | 2020-04-10 | 中电科特种飞机系统工程有限公司 | Training sample library construction method, device and medium based on air-to-ground small and slow target |
| CN111144398A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Target detection method, target detection device, computer equipment and storage medium |
| CN111310775A (en) * | 2018-12-11 | 2020-06-19 | Tcl集团股份有限公司 | Data training method and device, terminal equipment and computer readable storage medium |
| CN111462094A (en) * | 2020-04-03 | 2020-07-28 | 联觉(深圳)科技有限公司 | PCBA component detection method and device and computer readable storage medium |
| WO2020151329A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技（深圳）有限公司 | Target detection based identification box determining method and device and terminal equipment |
| CN112001375A (en) * | 2020-10-29 | 2020-11-27 | 成都睿沿科技有限公司 | Flame detection method and device, electronic equipment and storage medium |
| CN112256906A (en) * | 2020-10-23 | 2021-01-22 | 安徽启新明智科技有限公司 | Method, device and storage medium for marking annotation on display screen |
| CN112464785A (en) * | 2020-11-25 | 2021-03-09 | 浙江大华技术股份有限公司 | Target detection method and device, computer equipment and storage medium |
| CN112513935A (en) * | 2018-08-10 | 2021-03-16 | 奥林巴斯株式会社 | Image processing method and image processing apparatus |
| CN112949614A (en) * | 2021-04-29 | 2021-06-11 | 成都市威虎科技有限公司 | Face detection method and device for automatically allocating candidate areas and electronic equipment |
| CN113239975A (en) * | 2021-04-21 | 2021-08-10 | 洛阳青鸟网络科技有限公司 | Target detection method and device based on neural network |
| CN113333321A (en) * | 2021-05-11 | 2021-09-03 | 北京若贝特智能机器人科技有限公司 | Automatic identification and classification conveying method, system and device and storage medium |
| CN113496223A (en) * | 2020-03-19 | 2021-10-12 | 顺丰科技有限公司 | Method and device for establishing text region detection model |
| WO2022001501A1 (en) * | 2020-06-29 | 2022-01-06 | 华为技术有限公司 | Data annotation method and apparatus, and computer device and storage medium |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101271515A (en) * | 2007-03-21 | 2008-09-24 | 株式会社理光 | Image detection device capable of recognizing multi-angle objective |
| US20150117760A1 (en) * | 2013-10-30 | 2015-04-30 | Nec Laboratories America, Inc. | Regionlets with Shift Invariant Neural Patterns for Object Detection |
| WO2015078185A1 (en) * | 2013-11-29 | 2015-06-04 | 华为技术有限公司 | Convolutional neural network and target object detection method based on same |
| CN106250812A (en) * | 2016-07-15 | 2016-12-21 | 汤平 | A kind of model recognizing method based on quick R CNN deep neural network |
| CN106647758A (en) * | 2016-12-27 | 2017-05-10 | 深圳市盛世智能装备有限公司 | Target object detection method and device and automatic guiding vehicle following method |
| CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
| CN106874894A (en) * | 2017-03-28 | 2017-06-20 | 电子科技大学 | A kind of human body target detection method based on the full convolutional neural networks in region |
| CN104299012B (en) * | 2014-10-28 | 2017-06-30 | 银河水滴科技（北京）有限公司 | A kind of gait recognition method based on deep learning |
| CN106919978A (en) * | 2017-01-18 | 2017-07-04 | 西南交通大学 | A kind of high ferro contact net support meanss parts recognition detection method |
-
2017
- 2017-12-29 CN CN201711484723.3A patent/CN108121986B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101271515A (en) * | 2007-03-21 | 2008-09-24 | 株式会社理光 | Image detection device capable of recognizing multi-angle objective |
| US20150117760A1 (en) * | 2013-10-30 | 2015-04-30 | Nec Laboratories America, Inc. | Regionlets with Shift Invariant Neural Patterns for Object Detection |
| WO2015078185A1 (en) * | 2013-11-29 | 2015-06-04 | 华为技术有限公司 | Convolutional neural network and target object detection method based on same |
| CN104299012B (en) * | 2014-10-28 | 2017-06-30 | 银河水滴科技（北京）有限公司 | A kind of gait recognition method based on deep learning |
| CN106250812A (en) * | 2016-07-15 | 2016-12-21 | 汤平 | A kind of model recognizing method based on quick R CNN deep neural network |
| CN106647758A (en) * | 2016-12-27 | 2017-05-10 | 深圳市盛世智能装备有限公司 | Target object detection method and device and automatic guiding vehicle following method |
| CN106919978A (en) * | 2017-01-18 | 2017-07-04 | 西南交通大学 | A kind of high ferro contact net support meanss parts recognition detection method |
| CN106845430A (en) * | 2017-02-06 | 2017-06-13 | 东华大学 | Pedestrian detection and tracking based on acceleration region convolutional neural networks |
| CN106874894A (en) * | 2017-03-28 | 2017-06-20 | 电子科技大学 | A kind of human body target detection method based on the full convolutional neural networks in region |
Non-Patent Citations (4)
| Title |
|---|
| CHAOYUE CHEN ET AL: "Learning oriented region-based convolutional neural networks for building detection in satellite remote sensing images", 《REMOTE SENSING AND SPATIAL INFORMATION SCIENCES》 * |
| CHENCHEN ZHU ET AL: "CMS-RCNN:contextual multi-scale region-based CNN for unconstrained face detection", 《DEEP LEARNING FOR BIOMETRICS》 * |
| 杨名等: "甚高速区域卷积神经网络的船舶视频检测方法", 《北京邮电大学学报》 * |
| 胡鹏等: "基于区域卷积神经网络的多目标跟踪算法", 《西南科技大学学报》 * |
Cited By (46)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109101966A (en) * | 2018-06-08 | 2018-12-28 | 中国科学院宁波材料技术与工程研究所 | Workpiece identification positioning and posture estimation system and method based on deep learning |
| CN109101966B (en) * | 2018-06-08 | 2022-03-08 | 中国科学院宁波材料技术与工程研究所 | Workpiece recognition positioning and attitude estimation system and method based on deep learning |
| CN110619255A (en) * | 2018-06-19 | 2019-12-27 | 杭州海康威视数字技术股份有限公司 | Target detection method and device |
| CN110619255B (en) * | 2018-06-19 | 2022-08-26 | 杭州海康威视数字技术股份有限公司 | Target detection method and device |
| CN108921099A (en) * | 2018-07-03 | 2018-11-30 | 常州大学 | Moving ship object detection method in a kind of navigation channel based on deep learning |
| CN109063740A (en) * | 2018-07-05 | 2018-12-21 | 高镜尧 | The detection model of ultrasonic image common-denominator target constructs and detection method, device |
| CN110796154B (en) * | 2018-08-03 | 2023-03-24 | 华为云计算技术有限公司 | Method, device and equipment for training object detection model |
| CN110796154A (en) * | 2018-08-03 | 2020-02-14 | 华为技术有限公司 | Method, device and equipment for training object detection model |
| US11605211B2 (en) | 2018-08-03 | 2023-03-14 | Huawei Cloud Computing Technologies Co., Ltd. | Object detection model training method and apparatus, and device |
| CN112513935A (en) * | 2018-08-10 | 2021-03-16 | 奥林巴斯株式会社 | Image processing method and image processing apparatus |
| CN110837760B (en) * | 2018-08-17 | 2022-10-14 | 北京四维图新科技股份有限公司 | Target detection method, training method and device for target detection |
| CN110837760A (en) * | 2018-08-17 | 2020-02-25 | 北京四维图新科技股份有限公司 | Target detection method, training method and device for target detection |
| CN109376619A (en) * | 2018-09-30 | 2019-02-22 | 中国人民解放军陆军军医大学 | A kind of cell detection method |
| CN109376619B (en) * | 2018-09-30 | 2021-10-15 | 中国人民解放军陆军军医大学 | Cell detection method |
| CN109359683A (en) * | 2018-10-15 | 2019-02-19 | 百度在线网络技术（北京）有限公司 | Object detection method, device, terminal and computer readable storage medium |
| CN111144398A (en) * | 2018-11-02 | 2020-05-12 | 银河水滴科技（北京）有限公司 | Target detection method, target detection device, computer equipment and storage medium |
| CN109583445B (en) * | 2018-11-26 | 2024-08-02 | 平安科技（深圳）有限公司 | Text image correction processing method, device, equipment and storage medium |
| CN109583445A (en) * | 2018-11-26 | 2019-04-05 | 平安科技（深圳）有限公司 | Character image correction processing method, device, equipment and storage medium |
| CN109583396A (en) * | 2018-12-05 | 2019-04-05 | 广东亿迅科技有限公司 | A kind of region prevention method, system and terminal based on CNN two stages human testing |
| CN111310775B (en) * | 2018-12-11 | 2023-08-25 | Tcl科技集团股份有限公司 | Data training method, device, terminal equipment and computer readable storage medium |
| CN111310775A (en) * | 2018-12-11 | 2020-06-19 | Tcl集团股份有限公司 | Data training method and device, terminal equipment and computer readable storage medium |
| CN109784385A (en) * | 2018-12-29 | 2019-05-21 | 广州海昇计算机科技有限公司 | A kind of commodity automatic identifying method, system, device and storage medium |
| CN109934088A (en) * | 2019-01-10 | 2019-06-25 | 海南大学 | Sea ship discrimination method based on deep learning |
| CN109886998A (en) * | 2019-01-23 | 2019-06-14 | 平安科技（深圳）有限公司 | Multi-object tracking method, device, computer installation and computer storage medium |
| WO2020151166A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技（深圳）有限公司 | Multi-target tracking method and device, computer device and readable storage medium |
| WO2020151329A1 (en) * | 2019-01-23 | 2020-07-30 | 平安科技（深圳）有限公司 | Target detection based identification box determining method and device and terminal equipment |
| CN110288082A (en) * | 2019-06-05 | 2019-09-27 | 北京字节跳动网络技术有限公司 | Convolutional neural networks model training method, device and computer readable storage medium |
| CN110428357A (en) * | 2019-08-09 | 2019-11-08 | 厦门美图之家科技有限公司 | The detection method of watermark, device, electronic equipment and storage medium in image |
| CN110443244A (en) * | 2019-08-12 | 2019-11-12 | 深圳市捷顺科技实业股份有限公司 | A kind of method and relevant apparatus of graphics process |
| CN110443244B (en) * | 2019-08-12 | 2023-12-05 | 深圳市捷顺科技实业股份有限公司 | Graphics processing method and related device |
| CN110599456A (en) * | 2019-08-13 | 2019-12-20 | 杭州智团信息技术有限公司 | Method for extracting specific region of medical image |
| CN110717905B (en) * | 2019-09-30 | 2022-07-05 | 上海联影智能医疗科技有限公司 | Brain image detection method, computer device, and storage medium |
| CN110717905A (en) * | 2019-09-30 | 2020-01-21 | 上海联影智能医疗科技有限公司 | Brain image detection method, computer device, and storage medium |
| CN110807431A (en) * | 2019-11-06 | 2020-02-18 | 上海眼控科技股份有限公司 | Object positioning method and device, electronic equipment and storage medium |
| CN110991531A (en) * | 2019-12-02 | 2020-04-10 | 中电科特种飞机系统工程有限公司 | Training sample library construction method, device and medium based on air-to-ground small and slow target |
| CN113496223A (en) * | 2020-03-19 | 2021-10-12 | 顺丰科技有限公司 | Method and device for establishing text region detection model |
| CN111462094A (en) * | 2020-04-03 | 2020-07-28 | 联觉(深圳)科技有限公司 | PCBA component detection method and device and computer readable storage medium |
| WO2022001501A1 (en) * | 2020-06-29 | 2022-01-06 | 华为技术有限公司 | Data annotation method and apparatus, and computer device and storage medium |
| CN112256906A (en) * | 2020-10-23 | 2021-01-22 | 安徽启新明智科技有限公司 | Method, device and storage medium for marking annotation on display screen |
| CN112001375A (en) * | 2020-10-29 | 2020-11-27 | 成都睿沿科技有限公司 | Flame detection method and device, electronic equipment and storage medium |
| CN112464785A (en) * | 2020-11-25 | 2021-03-09 | 浙江大华技术股份有限公司 | Target detection method and device, computer equipment and storage medium |
| CN112464785B (en) * | 2020-11-25 | 2024-08-09 | 浙江大华技术股份有限公司 | Target detection method, device, computer equipment and storage medium |
| CN113239975B (en) * | 2021-04-21 | 2022-12-20 | 国网甘肃省电力公司白银供电公司 | Target detection method and device based on neural network |
| CN113239975A (en) * | 2021-04-21 | 2021-08-10 | 洛阳青鸟网络科技有限公司 | Target detection method and device based on neural network |
| CN112949614A (en) * | 2021-04-29 | 2021-06-11 | 成都市威虎科技有限公司 | Face detection method and device for automatically allocating candidate areas and electronic equipment |
| CN113333321A (en) * | 2021-05-11 | 2021-09-03 | 北京若贝特智能机器人科技有限公司 | Automatic identification and classification conveying method, system and device and storage medium |
Also Published As
| Publication number | Publication date |
|---|---|
| CN108121986B (en) | 2019-12-17 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108121986A (en) | Object detection method and device, computer installation and computer readable storage medium | |
| CN109886998A (en) | Multi-object tracking method, device, computer installation and computer storage medium | |
| CN104424634B (en) | Object tracking method and device | |
| CN108470354A (en) | Video target tracking method, device and realization device | |
| Shen et al. | Exemplar-based human action pose correction and tagging | |
| US11574430B2 (en) | Method and system for creating animal type avatar using human face | |
| CN110472534A (en) | 3D object detection method, device, equipment and storage medium based on RGB-D data | |
| CN111185008B (en) | Method and apparatus for controlling virtual character in game | |
| CN112699832B (en) | Target detection method, device, equipment and storage medium | |
| WO2022062238A1 (en) | Football detection method and apparatus, and computer-readable storage medium and robot | |
| CN107341548A (en) | A kind of data processing method, device and electronic equipment | |
| CN106874906A (en) | A kind of binarization method of picture, device and terminal | |
| CN109902541A (en) | A kind of method and system of image recognition | |
| CN109614990A (en) | A kind of object detecting device | |
| CN107871143A (en) | Image-recognizing method and device, computer installation and computer-readable recording medium | |
| CN111915657A (en) | Point cloud registration method and device, electronic equipment and storage medium | |
| CN114972947B (en) | Depth scene text detection method and device based on fuzzy semantic modeling | |
| CN110287857A (en) | A kind of training method of characteristic point detection model | |
| CN108229658A (en) | The implementation method and device of object detector based on finite sample | |
| CN107895021B (en) | image recognition method and device, computer device and computer readable storage medium | |
| CN108256400A (en) | The method for detecting human face of SSD based on deep learning | |
| CN112837322A (en) | Image segmentation method and device, equipment and storage medium | |
| CN103295026B (en) | Based on the image classification method of space partial polymerization description vectors | |
| CN112749576B (en) | Image recognition method and device, computing equipment and computer storage medium | |
| CN109712146A (en) | A kind of EM multi-threshold image segmentation method and device based on histogram |
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 |