CN110069961A - A kind of object detecting method and device - Google Patents
A kind of object detecting method and device Download PDFInfo
- Publication number
- CN110069961A CN110069961A CN201810067225.7A CN201810067225A CN110069961A CN 110069961 A CN110069961 A CN 110069961A CN 201810067225 A CN201810067225 A CN 201810067225A CN 110069961 A CN110069961 A CN 110069961A
- Authority
- CN
- China
- Prior art keywords
- frame image
- current frame
- bounding box
- image
- testing result
- 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
- 238000000034 method Methods 0.000 title claims abstract description 42
- 238000001514 detection method Methods 0.000 claims abstract description 130
- 238000012360 testing method Methods 0.000 claims abstract description 84
- 230000006872 improvement Effects 0.000 claims abstract description 70
- 238000004422 calculation algorithm Methods 0.000 claims description 26
- 238000012937 correction Methods 0.000 claims description 24
- 239000000284 extract Substances 0.000 claims description 8
- 230000001629 suppression Effects 0.000 claims description 8
- 238000004590 computer program Methods 0.000 claims description 6
- 230000000717 retained effect Effects 0.000 claims description 6
- 238000005516 engineering process Methods 0.000 abstract description 12
- 238000010586 diagram Methods 0.000 description 11
- 230000006870 function Effects 0.000 description 9
- 238000004364 calculation method Methods 0.000 description 6
- 230000006854 communication Effects 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000004891 communication Methods 0.000 description 5
- 235000013399 edible fruits Nutrition 0.000 description 5
- 230000000694 effects Effects 0.000 description 5
- 238000007689 inspection Methods 0.000 description 5
- 238000000605 extraction Methods 0.000 description 4
- 230000008901 benefit Effects 0.000 description 3
- 238000013527 convolutional neural network Methods 0.000 description 3
- 238000012986 modification Methods 0.000 description 3
- 230000004048 modification Effects 0.000 description 3
- 230000003287 optical effect Effects 0.000 description 3
- 230000008569 process Effects 0.000 description 3
- 230000011218 segmentation Effects 0.000 description 3
- 230000005291 magnetic effect Effects 0.000 description 2
- 230000001537 neural effect Effects 0.000 description 2
- 239000004065 semiconductor Substances 0.000 description 2
- 238000006467 substitution reaction Methods 0.000 description 2
- FGUUSXIOTUKUDN-IBGZPJMESA-N C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 Chemical compound C1(=CC=CC=C1)N1C2=C(NC([C@H](C1)NC=1OC(=NN=1)C1=CC=CC=C1)=O)C=CC=C2 FGUUSXIOTUKUDN-IBGZPJMESA-N 0.000 description 1
- 229910000831 Steel Inorganic materials 0.000 description 1
- 238000013528 artificial neural network Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000018109 developmental process Effects 0.000 description 1
- 230000005611 electricity Effects 0.000 description 1
- 239000000835 fiber Substances 0.000 description 1
- 210000003127 knee Anatomy 0.000 description 1
- 239000004973 liquid crystal related substance Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 239000010959 steel Substances 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Computational Linguistics (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of object detecting method and devices, are related to field of computer technology.One specific embodiment of this method includes: to detect to current frame image, to obtain the intermediate detection result of current frame image；It is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image, to obtain the improvement testing result of current frame image；The object in current frame image is determined according to the improvement testing result.The embodiment can be improved the accuracy rate of object detection.
Description
Technical field
The present invention relates to field of computer technology more particularly to a kind of object detecting methods and device.
Background technique
Object detection refers to the technology identified to the object in image.With the development of depth learning technology, object
Detection algorithm evolves as the algorithm based on deep learning by traditional solution based on feature.In the prior art,
When identifying the object in video file using object detection algorithms, video file is usually decomposed into single-frame images, then use
Object detection algorithms detect the object in each single-frame images respectively.
In realizing process of the present invention, at least there are the following problems in the prior art for inventor's discovery: existing object inspection
Method of determining and calculating individually detects each frame image, continuity and relevance without considering frame image, so as to cause
Detection accuracy is lower.
Summary of the invention
In view of this, the embodiment of the present invention provides a kind of object detecting method and device, the standard of object detection can be improved
True rate.
To achieve the above object, according to an aspect of an embodiment of the present invention, a kind of object detecting method is provided, is wrapped
It includes:
Current frame image is detected, to obtain the intermediate detection result of current frame image；
It is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image, to obtain
The improvement testing result of current frame image；
The object in current frame image is determined according to the improvement testing result.
Optionally, current frame image is detected, to include: the step of obtaining the intermediate detection result of current frame image
Multiple candidate frames are generated based on current frame image；Wherein, the coordinate representation of the coverage area candidate frame of candidate frame；
Calculate separately the probability that each candidate frame is surrounded by object, the confidence level as each candidate frame；
Candidate frame using confidence level greater than confidence threshold value is as bounding box, to calculate object category and the person in servitude of each bounding box
Belong to probability；Wherein, the coordinate of bounding box is the coordinate of corresponding candidate frame；
Using the coordinate of each bounding box, object category and it is subordinate to probability as intermediate detection result.
Optionally, it is repaired according to intermediate detection result of the improvement testing result of previous frame image to current frame image
Just, to include: the step of obtaining the improvement testing result of current frame image
It calculates each in the intermediate detection result of each bounding box and current frame image in the improvement testing result of previous frame image
The Duplication of bounding box；
It will be greater than the person in servitude of bounding box in the intermediate detection result of current frame image corresponding to the Duplication of Duplication threshold value
Belong to probability multiplied by correction factor, is subordinate to probability to obtain the amendment of the bounding box；
It is subordinate to probability according to the coordinate of bounding box each in current frame image, object category and amendment and generates improvement detection knot
Fruit.
Optionally, probability is subordinate to according to the coordinate of bounding box each in current frame image, object category and amendment and generates improvement
The step of testing result includes:
Duplicate removal is carried out to the bounding box in current frame image based on non-maxima suppression algorithm；
The coordinate of each bounding box retained after duplicate removal, object category and amendment are subordinate to probability as improvement testing result.
Optionally, include: based on the step of current frame image generation multiple candidate frames
Calculate the characteristic pattern of current frame image；
Current frame image is divided into multiple grids；
Judge respectively each grid position whether the seat with the bounding box in the improvement testing result of previous frame image
It marks corresponding；
If corresponding, the candidate frame for corresponding to the first quantity of the grid is generated；
If not corresponding to, the candidate frame for corresponding to the second quantity of the grid is generated；Wherein, first quantity is greater than institute
State the second quantity.
Optionally, the step of calculating object category that each bounding box is surrounded and being subordinate to probability before, further includes:
The coordinate of each bounding box is modified by returning sorter network.
Optionally, current frame image is detected, to include: the step of obtaining the intermediate detection result of current frame image
Extract the foreground image of current frame image；
Object detection is carried out to the foreground image of current frame image, to obtain the intermediate detection result of current frame image.
Optionally, the step of extracting the prospect of current frame image include:
Judge current frame image whether be frame image sequence first frame image；
If so, initializing foreground extractor using current frame image, and using current frame image as current frame image
Foreground image；
If it is not, then updating the foreground extractor using current frame image, and extracted currently using the foreground extractor
The foreground image of frame image.
To achieve the above object, other side according to an embodiment of the present invention provides a kind of article detection device, packet
It includes:
Detection module, for being detected to current frame image, to obtain the intermediate detection result of current frame image；
Correction module, for according to the improvement testing result of previous frame image to the intermediate detection result of current frame image into
Row amendment, to obtain the improvement testing result of current frame image；
Object determining module, for determining the object in current frame image according to the improvement testing result.
Optionally, the detection module is also used to:
Multiple candidate frames are generated based on current frame image；Wherein, the coordinate representation of the coverage area candidate frame of candidate frame；
Calculate separately the probability that each candidate frame is surrounded by object, the confidence level as each candidate frame；
Candidate frame using confidence level greater than confidence threshold value is as bounding box, to calculate object category and the person in servitude of each bounding box
Belong to probability；Wherein, the coordinate of bounding box is the coordinate of corresponding candidate frame；
Using the coordinate of each bounding box, object category and it is subordinate to probability as intermediate detection result.
Optionally, the correction module is also used to:
It calculates each in the intermediate detection result of each bounding box and current frame image in the improvement testing result of previous frame image
The Duplication of bounding box；
It will be greater than the person in servitude of bounding box in the intermediate detection result of current frame image corresponding to the Duplication of Duplication threshold value
Belong to probability multiplied by correction factor, is subordinate to probability to obtain the amendment of the bounding box；
It is subordinate to probability according to the coordinate of bounding box each in current frame image, object category and amendment and generates improvement detection knot
Fruit.
Optionally, the correction module is also used to:
Duplicate removal is carried out to the bounding box in current frame image based on non-maxima suppression algorithm；
The coordinate of each bounding box retained after duplicate removal, object category and amendment are subordinate to probability as improvement testing result.
Optionally, the correction module is also used to:
Calculate the characteristic pattern of current frame image；
Current frame image is divided into multiple grids；
Judge respectively each grid position whether the seat with the bounding box in the improvement testing result of previous frame image
It marks corresponding；
If corresponding, the candidate frame for corresponding to the first quantity of the grid is generated；
If not corresponding to, the candidate frame for corresponding to the second quantity of the grid is generated；Wherein, first quantity is greater than institute
State the second quantity.
Optionally, the correction module is also used to:
The coordinate of each bounding box is modified by returning sorter network.
Optionally, the detection module is also used to:
Extract the foreground image of current frame image；
Object detection is carried out to the foreground image of current frame image, to obtain the intermediate detection result of current frame image.
Optionally, the detection module is also used to:
Judge current frame image whether be frame image sequence first frame image；
If so, initializing foreground extractor using current frame image, and using current frame image as current frame image
Foreground image；
If it is not, then updating the foreground extractor using current frame image, and extracted currently using the foreground extractor
The foreground image of frame image.
To achieve the above object, another aspect according to an embodiment of the present invention provides a kind of object detection electronic and sets
It is standby, comprising:
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processing
Device is at least realized:
Current frame image is detected, to obtain the intermediate detection result of current frame image；
It is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image, to obtain
The improvement testing result of current frame image；
The object in current frame image is determined according to the improvement testing result.
To achieve the above object, another aspect according to an embodiment of the present invention provides a kind of computer-readable medium,
On be stored with computer program, at least realized when described program is executed by processor:
Current frame image is detected, to obtain the intermediate detection result of current frame image；
It is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image, to obtain
The improvement testing result of current frame image；
The object in current frame image is determined according to the improvement testing result.
One embodiment in foregoing invention has the following advantages that or the utility model has the advantages that because uses according to previous frame image
Improvement testing result technological means that the intermediate detection result of current frame image is modified, utilize image and object of which movement
Continuity testing result is improved, asked to solve the existing lower technology of object detection technology Detection accuracy
Topic has reached the technical effect for improving Detection accuracy.
Further effect possessed by above-mentioned non-usual optional way adds hereinafter in conjunction with specific embodiment
With explanation.
Detailed description of the invention
Attached drawing for a better understanding of the present invention, does not constitute an undue limitation on the present invention.Wherein:
Fig. 1 is the schematic diagram of the key step of object detecting method according to an embodiment of the present invention；
Fig. 2 is the schematic diagram of the main modular of article detection device according to an embodiment of the present invention；
Fig. 3 is that the embodiment of the present invention can be applied to exemplary system architecture figure therein；
Fig. 4 is adapted for the structural representation of the computer system for the terminal device or server of realizing the embodiment of the present invention
Figure.
Specific embodiment
Below in conjunction with attached drawing, an exemplary embodiment of the present invention will be described, including the various of the embodiment of the present invention
Details should think them only exemplary to help understanding.Therefore, those of ordinary skill in the art should recognize
It arrives, it can be with various changes and modifications are made to the embodiments described herein, without departing from scope and spirit of the present invention.Together
Sample, for clarity and conciseness, descriptions of well-known functions and structures are omitted from the following description.
Fig. 1 is the schematic diagram of the key step of object detecting method according to an embodiment of the present invention.
As shown in Figure 1, object detecting method provided in an embodiment of the present invention, comprising:
S100 detects current frame image, to obtain the intermediate detection result of current frame image.Wherein, among this
Testing result refers to the result that the object detection algorithms based on routine obtain after directly detecting.
S101 is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image,
To obtain the improvement testing result of current frame image.In this step, modified content refers mainly to detect in intermediate detection result
To object belong to the probability of a certain classification.Due to the movement of frame image and object be all it is continuous, consecutive frame image it
Between object space, size in the picture should be closer to；It, can be by comparing in present frame based on this thought
Between testing result and former frame improvement testing result (i.e. final detection result), identify close object (in the algorithm with
Bounding box is indicated), and the probability that this type objects in present frame are a certain classification is improved, to improve the accurate of detection
Degree.
S102 determines the object in current frame image according to the improvement testing result.This step is true using testing result
The object in image is determined, to provide accurate data for functions such as further tracking, identification, movement statistics.
From the above it can be seen that method provided in this embodiment is examined because using according to the improvement of previous frame image
The technological means that result is modified the intermediate detection result of current frame image is surveyed, the continuity of image and object of which movement is utilized
Testing result is improved, to solve the lower technical problem of existing object detection technology Detection accuracy, is reached
Improve the technical effect of Detection accuracy.
In some alternative embodiments, S100 detects current frame image, to obtain the centre of current frame image
The step of testing result includes:
Multiple candidate frames are generated based on current frame image.Candidate frame is the frame for carrying out tentative prediction to object space
Body, it is generally rectangular.
Calculate separately the probability that each candidate frame is surrounded by object, the confidence level as each candidate frame.This step is using existing
The algorithm of technology is realized, such as Faster RCNN (Faster Region-based Convolutional Neural
Nteworks, fast area convolutional neural networks), SSD (Single Shot MultiBox Detector, be based on more detection blocks
Single sweep operation detector).
Candidate frame using confidence level greater than confidence threshold value is as bounding box, by returning sorter network to each bounding box
Coordinate is modified.It returns sorter network and belongs to the prior art, bounding box can be made more accurately to surround object after being corrected
Edge.
It calculates the object category of each bounding box and is subordinate to probability.The calculating of this step equally uses the calculation of the above-mentioned prior art
Method is realized.Object category refers in the bounding box most possibly comprising which type objects, is subordinate to probability and is then used to indicate the encirclement
Box includes the probability of this type objects；For single bounding box, calculating its object category and being subordinate to the process of probability includes: point
Not Ji Suan the object in bounding box belong to the probability of multiple pre-set categories, and it is maximum probability is general as being subordinate to for bounding box
Rate, using the corresponding pre-set categories of maximum probability as the object category of bounding box.
Using the coordinate of each bounding box, object category and it is subordinate to probability as intermediate detection result.
S101 is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image,
To include: the step of obtaining the improvement testing result of current frame image
It calculates each in the intermediate detection result of each bounding box and current frame image in the improvement testing result of previous frame image
The Duplication of bounding box.Duplication is used to indicate the degree of two bounding boxs overlapping, and Duplication is higher, then illustrates two bounding boxs
Position, area it is closer, due to the continuity of time, it can be said that the two bright bounding boxs have very maximum probability to indicate identical
Object.
It will be greater than the person in servitude of bounding box in the intermediate detection result of current frame image corresponding to the Duplication of Duplication threshold value
Belong to probability multiplied by correction factor, is subordinate to probability to obtain the amendment of the bounding box.Correction factor is a real number greater than 1,
So that the amendment obtained after calculating is subordinate to probability and is subordinate to probability greater than script.
Duplicate removal is carried out to the bounding box in current frame image based on non-maxima suppression algorithm.Non-maxima suppression algorithm is
A kind of effective ways obtaining local maximum can be selected optimal after the algorithm duplicate removal from multiple bounding boxs of overlapping
Reservation as a result.
The coordinate of each bounding box retained after duplicate removal, object category and amendment are subordinate to probability as improvement testing result.
In above-described embodiment, when carrying out object detection to current frame image, pass through the bounding box position in previous frame image
It sets and the probability that is subordinate to of bounding box in current frame image is modified, improve identification using the continuity of image and object of which movement
The accuracy rate that algorithm identifies object category.
In some alternative embodiments, above-mentioned the step of generating multiple candidate frames based on current frame image, includes:
Calculate the characteristic pattern of current frame image.The calculating of characteristic pattern uses convolution operation or any other in the prior art
Algorithm is realized.
Current frame image is divided into multiple grids.Grid is usually evenly dividing, such as can be uniform by whole image
It is divided into the grid etc. of 3*3 or 9*9
Judge respectively each grid position whether the seat with the bounding box in the improvement testing result of previous frame image
It marks corresponding.In this step, if the center (usually geometric center) of bounding box is fallen within the scope of grid, determine the grid with
The bounding box is corresponding.
If corresponding, the candidate frame for corresponding to the first quantity of the grid is generated.If not corresponding to, generates and correspond to the net
The candidate frame of second quantity of lattice；Wherein, first quantity is greater than second quantity, the specific value of the two, Yi Ji
The set-up mode of candidate frame can be determined according to actual needs when the first quantity and the second quantity.
The present embodiment is improved by the way that more candidate frames are arranged at the position of bounding box in the testing result of former frame
Missing inspection is reduced so as to ensure that examined object has bigger probability to be surrounded by candidate frame for the number of object prediction
Rate.Simultaneously as increasing only the candidate frame quantity at grid corresponding with box position is surrounded, therefore unnecessary meter is saved
Calculation amount reduces the demand for computing resource.
In some alternative embodiments, S100 detects current frame image, to obtain the centre of current frame image
The step of testing result includes:
Extract the foreground image of current frame image.When extracting, judge whether current frame image is frame image sequence
First frame image；If so, initializing foreground extractor using current frame image, and using current frame image as present frame figure
The foreground image of picture；If it is not, then updating the foreground extractor using current frame image, and extracted using the foreground extractor
The foreground image of current frame image.
Object detection is carried out to the foreground image of current frame image, to obtain the intermediate detection result of current frame image.
The foreground image that the present embodiment method extracts frame image is detected, to reduce background parts ambient noise pair
It is interfered caused by testing result, improves the accuracy rate of detection.
The object detection that method in the embodiment of the present invention is suitable for carrying out consecutive image data.For example, can be used for pair
The object detection of video file can be used for the object detection for implementing monitoring image.Below based on to video file
The implementation of object detection overall description this method.
Step 1: enabling t=1, read the 1st frame image in video, be denoted as I1.Use I1Initialize foreground extractor；Due to
One frame image can not obtain prospect, the prospect F of first frame using foreground/background segmentation algorithm1=I1.Wherein, foreground extractor is adopted
The algorithm of the prospect of extraction can rely on the prior art and realize, such as can be existed using P.KadewTraKuPong and R.Bowden
The foreground/background segmentation algorithm (BackgroundSubtractorMOG) based on mixed Gauss model proposed in 2001；
The improvement foreground/background segmentation algorithm based on mixed Gauss model that Z.Zivkovic was proposed in 2004
(BackgroundSubtractorMOG2)；The probability prospect that Andrew_B.Godbehere et al. was proposed in 2012 is estimated to calculate
Method (BackgroundSubtractorGMG) etc..
Step 2: I is detected using object detecting method1In object, record object category C1And position B1.For first frame
For image, be not present previous frame image, thus the object detecting method that uses of step 2 can be used it is in the prior art any
Method；The object detecting method of mainstream is based on convolutional neural networks (CNN, Convolutional Neural at present
), such as previously mentioned Faster RCNN and SSD etc. Network.
Step 3: enabling t=t+1, read the t frame image in video, be denoted as It.Use ItForeground extractor is updated, before use
The prospect F of scape extractor extraction t frame imaget。
Step 4: the prospect F of t frame imagetAnd the object space B of t-1 framet-1As input, after acting on improvement
Object detection module, obtain the testing result of t frame image, record the object category C detectedt, object belong to the category
Probability PtAnd position Bt；Wherein, Ct、BtAnd PtIt can respectively include the corresponding numerical value of multiple groups bounding box.Improved object inspection
The detailed technology scheme of module is surveyed referring to hereinafter.
Step 5: if t has been the last frame of video, algorithm terminates, and exports the time series { C of objectt, Bt, Pt, t
∈ [1, N], wherein N is the totalframes that video file is read.Otherwise, step (3) are skipped to；It is carried out when for real-time image data
When detection, step 5 can be modified are as follows: according to the testing result renewal time sequence { C to t frame imaget, Bt, Pt, and skip to
Step (3).
In above-mentioned steps 4, sub-step performed by improved object detection module includes:
Step a: present frame foreground image F is extracted using convolution operationtCharacteristic pattern.The method of the present embodiment is based on convolution
The related algorithm of neural network, therefore realized when extracting characteristic pattern using convolution operation.
Step b: being divided into image uniform several grid, each grid is generated several centered on the grid
A candidate frame, k-th of candidate frame that a grid of note (i, j) generates areThe quantity for the candidate frame that each grid generates is by upper
The testing result B of one framet-1It determines: with Bt-1The candidate frame of coincidence is more.Wherein, " coincidence " is meant that Bt-1In packet
It encloses the center box (boundingbox) to fall in grid, bounding box is generally rectangular, and center is usually its geometric center.Work as Bt-1
In bounding box when having multiple, then for all generating more candidate frames in the presence of the grid for being overlapped bounding box.
For tentatively predicting the object in image, the generating mode of the candidate frame in the present embodiment is candidate frame,
Centered on net center of a lattice, generate that multiple areas are certain, length-width ratio is the rectangle of fixed proportion, such as area can be generated and be
10, length-width ratio is respectively 3 rectangle frames of 1:3,1:1 and 3:1 as candidate frame.According to the method in step b, if grid is with before
Bounding box in one frame image detection result is overlapped, then generates more candidate frames for the grid, such as can be generated
Additional one group of area is 15, the identical candidate frame of ratio；Also the candidate frame of different length-width ratios can be generated；Or it can also be by two kinds
Mode combines.By generating more candidate frames at grid, the detection range at grid can be increased, to avoid the occurrence of leakage
The problem of inspection；It, can be with simultaneously as be purposefully to increase candidate frame, therefore avoid additionally increasing more computation burdens
Improve the detection efficiency of algorithm.
Step c: for each candidate frameThe feature of corresponding position in characteristic pattern is extracted as box feature, by returning
Return network, judges that candidate frame surrounds the probability of the box of object, be denoted as confidence level.Wherein, " corresponding position " refers to candidate frame
Area encompassed corresponding image-region in characteristic pattern.
Step d: retain the candidate frame that confidence level is greater than preset confidence threshold value, referred to as bounding box, remember candidate frame
Corresponding bounding box isIt should be noted that candidate frame and bounding box are generally rectangular, therefore the seat of angle steel joint can be passed through
Mark or center point coordinate and its length and the modes such as width record their position.
Step e: by returning sorter network, to bounding boxIt is returned and is classified, correct the coordinate of bounding box, and
Obtain the object category and be subordinate to probability that bounding box is surrounded；Remember that the bounding box coordinate after returning isObject category is denoted asBounding boxIncluding objectProbability be denoted as
Step f: according to the detection bounding box B of previous framet-1The result of amendment step e.It calculatesWith Bt-1Friendship and ratio,
For handing over and than greater than Duplication threshold valueBeing modified to it includes that object probability isWherein Duplication threshold value
For the real number less than 1, coefficient b is the real number greater than 1, and the preferable value range of the two is 0.4 < a < 1,1 <b < 1.5.In this step
To hand over and compare the Duplication to indicate two bounding boxs, hands over and the calculation of ratio is removed using the intersection area of two bounding boxs
With their union area.WhenThere is multiple or Bt-1In bounding box when having multiple, can be respectively with eachWith Bt-1
In each bounding box calculate separately friendship and compare, and calculated result is selected to be greater than Duplication threshold valueAccording to above-mentioned formula to it
Corresponding includes that object probability is modified；Whole can also first be calculatedUnion and Bt-1It is middle whole bounding box and
Then collection calculates the friendship of two unions and ratio, if calculated result is greater than Duplication threshold value, to eachCorresponding includes object
Probability is modified；When using different calculations, the value of Duplication threshold value should also be as being adaptively adjusted.
In above-mentioned steps f, by introducing the object detection result of previous frame image, to the object detection of current frame image
As a result the probability that middle bounding box surrounds object is corrected.This amendment from video file based on the idea that due to mentioning
The frame image taken is continuously (can to extract each frame image according to computing capability at the extraction, or equally spaced in time
Extract image), thus the change in location of object in the picture should also be as be it is continuous, so can be according to more continuous two
The testing result of frame increases the identical probability of object type in two bounding boxs of position close (i.e. bounding box hand over and relatively high)
Add, to improve the accuracy rate of testing result.
G. basisAndThe high bounding box of registration is excluded using non-maxima suppression algorithm, output meets condition
'sAndThe C separately constitutedt、BtAnd PtTesting result as present frame (t frame).
In summary as can be seen that the method in the embodiment of the present invention is made that object detection technology changes from the following aspect
Into:
1. being modified, taking full advantage of to the testing result of current frame image using the testing result of previous frame image
Relevance between sequential frame image, to improve the accuracy rate of detection；
2. corresponding to the bounding box of former frame when carrying out candidate frame generation using the testing result of previous frame image
Grid generates more candidate frames, on the one hand can save computing resource under the premise of guaranteeing accuracy rate, on the other hand can be with
The case where avoiding missing inspection；
3. extraction prospect is detected before detecting to frame image, to reduce ambient noise to testing result
Interference.
Fig. 2 is the schematic diagram of the main modular of article detection device according to an embodiment of the present invention.
As shown in Fig. 2, the article detection device 200 provided according to embodiments of the present invention, comprising:
Detection module 201, for being detected to current frame image, to obtain the intermediate detection result of current frame image；
Wherein, which refers to the result that the object detection algorithms based on routine obtain after directly detecting.
Correction module 202, for the intermediate detection knot according to the improvement testing result of previous frame image to current frame image
Fruit is modified, to obtain the improvement testing result of current frame image；Modified content refers mainly to detect in intermediate detection result
Obtained object belongs to the probability of a certain classification.
Object determining module 203, for determining the object in current frame image according to the improvement testing result.
From the above it can be seen that device provided in this embodiment is examined because using according to the improvement of previous frame image
The technological means that result is modified the intermediate detection result of current frame image is surveyed, the continuity of image and object of which movement is utilized
Testing result is improved, to solve the lower technical problem of existing object detection technology Detection accuracy, is reached
Improve the technical effect of Detection accuracy.
In some alternative embodiments, the detection module 201 is also used to:
Multiple candidate frames are generated based on current frame image；
Calculate separately the probability that each candidate frame is surrounded by object, the confidence level as each candidate frame；
Confidence level is greater than the candidate frame of confidence threshold value as bounding box；
It calculates the object category of each bounding box and is subordinate to probability；
Using the coordinate of each bounding box, object category and it is subordinate to probability as intermediate detection result.
In some alternative embodiments, the correction module 202 is also used to:
It calculates each in the intermediate detection result of each bounding box and current frame image in the improvement testing result of previous frame image
The Duplication of bounding box；
It will be greater than the person in servitude of bounding box in the intermediate detection result of current frame image corresponding to the Duplication of Duplication threshold value
Belong to probability multiplied by correction factor, is subordinate to probability to obtain the amendment of the bounding box；
It is subordinate to probability according to the coordinate of bounding box each in current frame image, object category and amendment and generates improvement detection knot
Fruit.
In some alternative embodiments, the correction module 202 is also used to:
Duplicate removal is carried out to the bounding box in current frame image based on non-maxima suppression algorithm；
The coordinate of each bounding box retained after duplicate removal, object category and amendment are subordinate to probability as improvement testing result.
In some alternative embodiments, the correction module 202 is also used to:
Calculate the characteristic pattern of current frame image；
Current frame image is divided into multiple grids；
Judge respectively each grid position whether the seat with the bounding box in the improvement testing result of previous frame image
It marks corresponding；
If corresponding, the candidate frame for corresponding to the first quantity of the grid is generated；
If not corresponding to, the candidate frame for corresponding to the second quantity of the grid is generated；Wherein, first quantity is greater than institute
State the second quantity.
In some alternative embodiments in the correction module 202 is also used to:
The coordinate of each bounding box is modified by returning sorter network.
In some alternative embodiments, the detection module 201 is also used to:
Extract the foreground image of current frame image；
Object detection is carried out to the foreground image of current frame image, to obtain the intermediate detection result of current frame image.
In some alternative embodiments, the detection module 201 is also used to:
Judge current frame image whether be frame image sequence first frame image；
If so, initializing foreground extractor using current frame image, and using current frame image as current frame image
Foreground image；
If it is not, then updating the foreground extractor using current frame image, and extracted currently using the foreground extractor
The foreground image of frame image.
Fig. 3 is shown can be using the object detecting method of the embodiment of the present invention or the exemplary system of article detection device
Framework 300.
As shown in figure 3, system architecture 300 may include terminal device 301,302,303, network 304 and server 305.
Network 304 between terminal device 301,302,303 and server 305 to provide the medium of communication link.Network 304 can be with
Including various connection types, such as wired, wireless communication link or fiber optic cables etc..
User can be used terminal device 301,302,303 and be interacted by network 304 with server 305, to receive or send out
Send message etc..
Terminal device 301,302,303 can be the various electronic equipments with display screen and supported web page browsing, packet
Include but be not limited to smart phone, tablet computer, pocket computer on knee and desktop computer etc..
Server 305 can be to provide the server of various services, such as carry out processing to image data and tie processing
Fruit feeds back to the background server of terminal device 301,302,303.
It should be noted that object detecting method provided by the embodiment of the present invention is generally executed by server 305, accordingly
Ground, article detection device are generally positioned in server 305.
It should be understood that the number of terminal device, network and server in Fig. 3 is only schematical.According to realization need
It wants, can have any number of terminal device, network and server.
According to an embodiment of the invention, the present invention also provides a kind of electronic equipment and a kind of readable storage medium storing program for executing.
Fig. 4 is adapted for the structural representation of the computer system for the terminal device or server of realizing the embodiment of the present invention
Figure.
Below with reference to Fig. 4, it illustrates the computer systems 400 for the terminal device for being suitable for being used to realize the embodiment of the present invention
Structural schematic diagram.Terminal device shown in Fig. 4 is only an example, function to the embodiment of the present invention and should not use model
Shroud carrys out any restrictions.
As shown in figure 4, computer system 400 includes central processing unit (CPU) 401, it can be read-only according to being stored in
Program in memory (ROM) 402 or be loaded into the program in random access storage device (RAM) 403 from storage section 408 and
Execute various movements appropriate and processing.In RAM 403, also it is stored with system 400 and operates required various programs and data.
CPU 401, ROM 402 and RAM 403 are connected with each other by bus 404.Input/output (I/O) interface 405 is also connected to always
Line 404.
I/O interface 405 is connected to lower component: the importation 406 including keyboard, mouse etc.；It is penetrated including such as cathode
The output par, c 407 of spool (CRT), liquid crystal display (LCD) etc. and loudspeaker etc.；Storage section 408 including hard disk etc.；
And the communications portion 409 of the network interface card including LAN card, modem etc..Communications portion 409 via such as because
The network of spy's net executes communication process.Driver 410 is also connected to I/O interface 405 as needed.Detachable media 411, such as
Disk, CD, magneto-optic disk, semiconductor memory etc. are mounted on as needed on driver 410, in order to read from thereon
Computer program be mounted into storage section 408 as needed.
Particularly, according to an embodiment of the invention, the process of the schematic diagram description of key step may be implemented as above
Computer software programs.For example, the embodiment of the present invention includes a kind of computer program product comprising being carried on computer can
The computer program on medium is read, which includes the program for executing method shown in the schematic diagram of key step
Code.In such embodiments, which can be downloaded and installed from network by communications portion 409, and/
Or it is mounted from detachable media 411.When the computer program is executed by central processing unit (CPU) 401, the present invention is executed
System in the above-mentioned function that limits.
It should be noted that computer-readable medium shown in the present invention can be computer-readable signal media or meter
Calculation machine readable storage medium storing program for executing either the two any combination.Computer readable storage medium for example can be --- but not
Be limited to --- electricity, magnetic, optical, electromagnetic, infrared ray or semiconductor system, device or device, or any above combination.Meter
The more specific example of calculation machine readable storage medium storing program for executing can include but is not limited to: have the electrical connection, just of one or more conducting wires
Taking formula computer disk, hard disk, random access storage device (RAM), read-only memory (ROM), erasable type may be programmed read-only storage
Device (EPROM or flash memory), optical fiber, portable compact disc read-only memory (CD-ROM), light storage device, magnetic memory device,
Or above-mentioned any appropriate combination.In the present invention, computer readable storage medium can be it is any include or storage journey
The tangible medium of sequence, the program can be commanded execution system, device or device use or in connection.And at this
In invention, computer-readable signal media may include in a base band or as carrier wave a part propagate data-signal,
Wherein carry computer-readable program code.The data-signal of this propagation can take various forms, including but unlimited
In electromagnetic signal, optical signal or above-mentioned any appropriate combination.Computer-readable signal media can also be that computer can
Any computer-readable medium other than storage medium is read, which can send, propagates or transmit and be used for
By the use of instruction execution system, device or device or program in connection.Include on computer-readable medium
Program code can transmit with any suitable medium, including but not limited to: wireless, electric wire, optical cable, RF etc. are above-mentioned
Any appropriate combination.
Flow chart and block diagram in attached drawing are illustrated according to the system of various embodiments of the invention, method and computer journey
The architecture, function and operation in the cards of sequence product.In this regard, each box in flowchart or block diagram can generation
A part of one module, program segment or code of table, a part of above-mentioned module, program segment or code include one or more
Executable instruction for implementing the specified logical function.It should also be noted that in some implementations as replacements, institute in box
The function of mark can also occur in a different order than that indicated in the drawings.For example, two boxes succeedingly indicated are practical
On can be basically executed in parallel, they can also be executed in the opposite order sometimes, and this depends on the function involved.Also it wants
It is noted that the combination of each box in block diagram or flow chart and the box in block diagram or flow chart, can use and execute rule
The dedicated hardware based systems of fixed functions or operations is realized, or can use the group of specialized hardware and computer instruction
It closes to realize.
Being described in module involved in the embodiment of the present invention can be realized by way of software, can also be by hard
The mode of part is realized.Described module also can be set in the processor, for example, can be described as: a kind of processor packet
Include detection module, correction module and object determining module.Wherein, the title of these modules is not constituted to this under certain conditions
The restriction of module itself, for example, correction module is also described as " for the improvement testing result pair according to previous frame image
The intermediate detection result of current frame image is modified, to obtain the module of the improvement testing result of current frame image ".
As on the other hand, the present invention also provides a kind of computer-readable medium, which be can be
Included in equipment described in above-described embodiment；It is also possible to individualism, and without in the supplying equipment.Above-mentioned calculating
Machine readable medium carries one or more program, when said one or multiple programs are executed by the equipment, makes
Obtaining the equipment includes:
Current frame image is detected, to obtain the intermediate detection result of current frame image；
It is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image, to obtain
The improvement testing result of current frame image；
The object in current frame image is determined according to the improvement testing result.
Technical solution according to an embodiment of the present invention, because using according to the improvement testing result of previous frame image to working as
The technological means that the intermediate detection result of prior image frame is modified, using the continuity of image and object of which movement to testing result
It improves, to solve the lower technical problem of existing object detection technology Detection accuracy, has reached raising detection
The technical effect of accuracy rate.
Above-mentioned specific embodiment, does not constitute a limitation on the scope of protection of the present invention.Those skilled in the art should be bright
It is white, design requirement and other factors are depended on, various modifications, combination, sub-portfolio and substitution can occur.It is any
Made modifications, equivalent substitutions and improvements etc. within the spirit and principles in the present invention, should be included in the scope of the present invention
Within.
Claims (18)
1. a kind of object detecting method characterized by comprising
Current frame image is detected, to obtain the intermediate detection result of current frame image；
It is modified according to intermediate detection result of the improvement testing result of previous frame image to current frame image, it is current to obtain
The improvement testing result of frame image；
The object in current frame image is determined according to the improvement testing result.
2. the method according to claim 1, wherein being detected to current frame image, to obtain present frame figure
The step of intermediate detection result of picture includes:
Multiple candidate frames are generated based on current frame image；Wherein, the coordinate representation of the coverage area candidate frame of candidate frame；
Calculate separately the probability that each candidate frame is surrounded by object, the confidence level as each candidate frame；
Candidate frame using confidence level greater than confidence threshold value is as bounding box, to calculate the object category of each bounding box and be subordinate to general
Rate；Wherein, the coordinate of bounding box is the coordinate of corresponding candidate frame；
Using the coordinate of each bounding box, object category and it is subordinate to probability as intermediate detection result.
3. according to the method described in claim 2, it is characterized in that, according to the improvement testing result of previous frame image to present frame
The intermediate detection result of image is modified, to include: the step of obtaining the improvement testing result of current frame image
It calculates in the improvement testing result of previous frame image and is respectively surrounded in the intermediate detection result of each bounding box and current frame image
The Duplication of box；
It is general to will be greater than being subordinate to for bounding box in the intermediate detection result of current frame image corresponding to the Duplication of Duplication threshold value
Rate is subordinate to probability multiplied by correction factor, to obtain the amendment of the bounding box；
It is subordinate to probability according to the coordinate of bounding box each in current frame image, object category and amendment and generates improvement testing result.
4. according to the method described in claim 3, it is characterized in that, according to coordinate, the object of bounding box each in current frame image
Classification and amendment be subordinate to probability generate improve testing result the step of include:
Duplicate removal is carried out to the bounding box in current frame image based on non-maxima suppression algorithm；
The coordinate of each bounding box retained after duplicate removal, object category and amendment are subordinate to probability as improvement testing result.
5. according to the method described in claim 2, it is characterized in that, the step of generating multiple candidate frames based on current frame image is wrapped
It includes:
Calculate the characteristic pattern of current frame image；
Current frame image is divided into multiple grids；
Judge respectively each grid position whether the coordinate phase with the bounding box in the improvement testing result of previous frame image
It is corresponding；
If corresponding, the candidate frame for corresponding to the first quantity of the grid is generated；
If not corresponding to, the candidate frame for corresponding to the second quantity of the grid is generated；Wherein, first quantity is greater than described the
Two quantity.
6. according to the method described in claim 2, it is characterized in that, calculating the object category and be subordinate to that each bounding box is surrounded
Before the step of probability, further includes:
The coordinate of each bounding box is modified by returning sorter network.
7. the method according to claim 1, wherein being detected to current frame image, to obtain present frame figure
The step of intermediate detection result of picture includes:
Extract the foreground image of current frame image；
Object detection is carried out to the foreground image of current frame image, to obtain the intermediate detection result of current frame image.
8. the method according to the description of claim 7 is characterized in that the step of extracting the prospect of current frame image includes:
Judge current frame image whether be frame image sequence first frame image；
If so, initializing foreground extractor using current frame image, and using current frame image as the prospect of current frame image
Image；
If it is not, then updating the foreground extractor using current frame image, and present frame figure is extracted using the foreground extractor
The foreground image of picture.
9. a kind of article detection device characterized by comprising
Detection module, for being detected to current frame image, to obtain the intermediate detection result of current frame image；
Correction module, for being repaired according to the improvement testing result of previous frame image to the intermediate detection result of current frame image
Just, to obtain the improvement testing result of current frame image；
Object determining module, for determining the object in current frame image according to the improvement testing result.
10. device according to claim 9, which is characterized in that the detection module is also used to:
Multiple candidate frames are generated based on current frame image；Wherein, the coordinate representation of the coverage area candidate frame of candidate frame；
Calculate separately the probability that each candidate frame is surrounded by object, the confidence level as each candidate frame；
Candidate frame using confidence level greater than confidence threshold value is as bounding box, to calculate the object category of each bounding box and be subordinate to general
Rate；Wherein, the coordinate of bounding box is the coordinate of corresponding candidate frame；
Using the coordinate of each bounding box, object category and it is subordinate to probability as intermediate detection result.
11. device according to claim 10, which is characterized in that the correction module is also used to:
It calculates in the improvement testing result of previous frame image and is respectively surrounded in the intermediate detection result of each bounding box and current frame image
The Duplication of box；
It is general to will be greater than being subordinate to for bounding box in the intermediate detection result of current frame image corresponding to the Duplication of Duplication threshold value
Rate is subordinate to probability multiplied by correction factor, to obtain the amendment of the bounding box；
It is subordinate to probability according to the coordinate of bounding box each in current frame image, object category and amendment and generates improvement testing result.
12. device according to claim 11, which is characterized in that the correction module is also used to:
Duplicate removal is carried out to the bounding box in current frame image based on non-maxima suppression algorithm；
The coordinate of each bounding box retained after duplicate removal, object category and amendment are subordinate to probability as improvement testing result.
13. device according to claim 10, which is characterized in that the correction module is also used to:
Calculate the characteristic pattern of current frame image；
Current frame image is divided into multiple grids；
Judge respectively each grid position whether the coordinate phase with the bounding box in the improvement testing result of previous frame image
It is corresponding；
If corresponding, the candidate frame for corresponding to the first quantity of the grid is generated；
If not corresponding to, the candidate frame for corresponding to the second quantity of the grid is generated；Wherein, first quantity is greater than described the
Two quantity.
14. device according to claim 10, which is characterized in that the correction module is also used to:
The coordinate of each bounding box is modified by returning sorter network.
15. device according to claim 9, which is characterized in that the detection module is also used to:
Extract the foreground image of current frame image；
Object detection is carried out to the foreground image of current frame image, to obtain the intermediate detection result of current frame image.
16. device according to claim 15, which is characterized in that the detection module is also used to:
Judge current frame image whether be frame image sequence first frame image；
If so, initializing foreground extractor using current frame image, and using current frame image as the prospect of current frame image
Image；
If it is not, then updating the foreground extractor using current frame image, and present frame figure is extracted using the foreground extractor
The foreground image of picture.
17. a kind of electronic equipment for object detection characterized by comprising
One or more processors；
Storage device, for storing one or more programs,
When one or more of programs are executed by one or more of processors, so that one or more of processors are real
Now such as method described in any one of claims 1-8.
18. a kind of computer-readable medium, is stored thereon with computer program, which is characterized in that described program is held by processor
Such as method described in any one of claims 1-8 is realized when row.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810067225.7A CN110069961B (en) | 2018-01-24 | Object detection method and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810067225.7A CN110069961B (en) | 2018-01-24 | Object detection method and device |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110069961A true CN110069961A (en) | 2019-07-30 |
| CN110069961B CN110069961B (en) | 2024-07-16 |
Family
ID=
Cited By (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110097114A (en) * | 2019-04-26 | 2019-08-06 | 新华三技术有限公司 | A kind of priori frame applied to neural network determines method and device |
| CN110706227A (en) * | 2019-10-14 | 2020-01-17 | 普联技术有限公司 | Article state detection method, system, terminal device and storage medium |
| CN110717933A (en) * | 2019-10-10 | 2020-01-21 | 北京百度网讯科技有限公司 | Post-processing method, device, equipment and medium for moving object missed detection |
| CN111402414A (en) * | 2020-03-10 | 2020-07-10 | 北京京东叁佰陆拾度电子商务有限公司 | Point cloud map construction method, device, equipment and storage medium |
| CN111507997A (en) * | 2020-04-22 | 2020-08-07 | 腾讯科技（深圳）有限公司 | Image segmentation method, device, equipment and computer storage medium |
| CN112949354A (en) * | 2019-12-10 | 2021-06-11 | 顺丰科技有限公司 | Method and device for detecting wearing of safety helmet, electronic equipment and computer-readable storage medium |
| CN113505733A (en) * | 2021-07-26 | 2021-10-15 | 浙江大华技术股份有限公司 | Behavior recognition method, behavior recognition device, storage medium and electronic device |
Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011238227A (en) * | 2010-05-07 | 2011-11-24 | Ricoh Co Ltd | Method and device for object detection |
| CN102760230A (en) * | 2012-06-19 | 2012-10-31 | 华中科技大学 | Flame detection method based on multi-dimensional time domain characteristics |
| CN104715249A (en) * | 2013-12-16 | 2015-06-17 | 株式会社理光 | Object tracking method and device |
| CN105825524A (en) * | 2016-03-10 | 2016-08-03 | 浙江生辉照明有限公司 | Target tracking method and apparatus |
| CN106886755A (en) * | 2017-01-19 | 2017-06-23 | 北京航空航天大学 | A kind of intersection vehicles system for detecting regulation violation based on Traffic Sign Recognition |
| CN107452025A (en) * | 2017-08-18 | 2017-12-08 | 成都通甲优博科技有限责任公司 | Method for tracking target, device and electronic equipment |
Patent Citations (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2011238227A (en) * | 2010-05-07 | 2011-11-24 | Ricoh Co Ltd | Method and device for object detection |
| CN102760230A (en) * | 2012-06-19 | 2012-10-31 | 华中科技大学 | Flame detection method based on multi-dimensional time domain characteristics |
| CN104715249A (en) * | 2013-12-16 | 2015-06-17 | 株式会社理光 | Object tracking method and device |
| CN105825524A (en) * | 2016-03-10 | 2016-08-03 | 浙江生辉照明有限公司 | Target tracking method and apparatus |
| CN106886755A (en) * | 2017-01-19 | 2017-06-23 | 北京航空航天大学 | A kind of intersection vehicles system for detecting regulation violation based on Traffic Sign Recognition |
| CN107452025A (en) * | 2017-08-18 | 2017-12-08 | 成都通甲优博科技有限责任公司 | Method for tracking target, device and electronic equipment |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110097114A (en) * | 2019-04-26 | 2019-08-06 | 新华三技术有限公司 | A kind of priori frame applied to neural network determines method and device |
| CN110717933A (en) * | 2019-10-10 | 2020-01-21 | 北京百度网讯科技有限公司 | Post-processing method, device, equipment and medium for moving object missed detection |
| CN110717933B (en) * | 2019-10-10 | 2023-02-07 | 阿波罗智能技术(北京)有限公司 | Post-processing method, device, equipment and medium for moving object missed detection |
| CN110706227A (en) * | 2019-10-14 | 2020-01-17 | 普联技术有限公司 | Article state detection method, system, terminal device and storage medium |
| CN110706227B (en) * | 2019-10-14 | 2022-07-05 | 普联技术有限公司 | Article state detection method, system, terminal device and storage medium |
| CN112949354A (en) * | 2019-12-10 | 2021-06-11 | 顺丰科技有限公司 | Method and device for detecting wearing of safety helmet, electronic equipment and computer-readable storage medium |
| CN111402414A (en) * | 2020-03-10 | 2020-07-10 | 北京京东叁佰陆拾度电子商务有限公司 | Point cloud map construction method, device, equipment and storage medium |
| CN111402414B (en) * | 2020-03-10 | 2024-05-24 | 北京京东叁佰陆拾度电子商务有限公司 | Point cloud map construction method, device, equipment and storage medium |
| CN111507997A (en) * | 2020-04-22 | 2020-08-07 | 腾讯科技（深圳）有限公司 | Image segmentation method, device, equipment and computer storage medium |
| CN113505733A (en) * | 2021-07-26 | 2021-10-15 | 浙江大华技术股份有限公司 | Behavior recognition method, behavior recognition device, storage medium and electronic device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109117831A (en) | The training method and device of object detection network | |
| US20200334830A1 (en) | Method, apparatus, and storage medium for processing video image | |
| CN108090916B (en) | Method and apparatus for tracking the targeted graphical in video | |
| CN107911753A (en) | Method and apparatus for adding digital watermarking in video | |
| US20230334890A1 (en) | Pedestrian re-identification method and device | |
| CN109934181A (en) | Text recognition method, device, equipment and computer-readable medium | |
| CN110348522A (en) | A kind of image detection recognition methods and system, electronic equipment, image classification network optimized approach and system | |
| CN109308490A (en) | Method and apparatus for generating information | |
| CN109948637A (en) | Object test equipment, method for checking object and computer-readable medium | |
| CN111191730B (en) | Method and system for detecting oversized image target oriented to embedded deep learning | |
| CN110335313A (en) | Audio collecting device localization method and device, method for distinguishing speek person and system | |
| CN108241853A (en) | A kind of video frequency monitoring method, system and terminal device | |
| CN110349158A (en) | A kind of method and apparatus handling point cloud data | |
| CN108182457A (en) | For generating the method and apparatus of information | |
| CN110348508A (en) | Examine the data checking method and its system, electronic equipment of exceptional value | |
| CN109901988A (en) | A kind of page elements localization method and device for automatic test | |
| CN115294332A (en) | Image processing method, device, equipment and storage medium | |
| CN109697392A (en) | Draw the method and device of target object thermodynamic chart | |
| CN110363137A (en) | Face datection Optimized model, method, system and its electronic equipment | |
| CN109993749A (en) | The method and apparatus for extracting target image | |
| CN109978855A (en) | A kind of method for detecting change of remote sensing image and device | |
| CN108596120A (en) | A kind of object detection method and device based on deep learning | |
| CN115439700B (en) | Image processing method and device and machine-readable storage medium | |
| CN110069961A (en) | A kind of object detecting method and device | |
| CN108062576B (en) | Method and apparatus for output data |
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 |