CN110532985A - Object detection method, apparatus and system - Google Patents
Object detection method, apparatus and system Download PDFInfo
- Publication number
- CN110532985A CN110532985A CN201910830745.3A CN201910830745A CN110532985A CN 110532985 A CN110532985 A CN 110532985A CN 201910830745 A CN201910830745 A CN 201910830745A CN 110532985 A CN110532985 A CN 110532985A
- Authority
- CN
- China
- Prior art keywords
- whole body
- target object
- detection block
- body detection
- upper area
- 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
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
-
- 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
- G06V10/44—Local feature extraction by analysis of parts of the pattern, e.g. by detecting edges, contours, loops, corners, strokes or intersections; Connectivity analysis, e.g. of connected components
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- 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)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Data Mining & Analysis (AREA)
- Multimedia (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- General Engineering & Computer Science (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
The present invention provides a kind of object detection methods, apparatus and system, are related to field of artificial intelligence, include image to be detected of target object including acquisition；Target detection is carried out to image to be detected by preset at least two anchor points frame, generates at least two initial detecting results of target object；Wherein, corresponding second of the anchor point frame for generating a kind of initial detecting as a result, at least two anchor point frames include the first anchor point frame of the local feature setting based on target object generic and the systemic features setting based on target object generic of every kind of anchor point frame；At least two initial detecting results are merged, the final detection result of target object is obtained.The present invention can effectively promote the accuracy rate of object detection results.
Description
Technical field
The present invention relates to field of artificial intelligence, more particularly, to a kind of object detection method, apparatus and system.
Background technique
Target detection technique is an important branch in computer vision field, and main purpose is to detect from image
The target of pre-set categories (pedestrian, vehicle, cat etc.) out, such as, the systemic features based on pedestrian detect that image is included
The position of pedestrian.
However, inventor it has been investigated that, be likely to occur in existing target detection technique because target to be detected is blocked
And the case where being difficult to detect by target, cause the accuracy rate of object detection results not high.For ease of understanding, it is with pedestrian detection
Example, in fact it could happen that the case where body of pedestrian is blocked by other objects (tables and chairs, shopping cart, estrade etc.), due to portion
Attend to anything else the missings of body characteristics, pedestrian detection technology is difficult to detect by the pedestrian at this time, to influence the accurate of pedestrian detection result
Rate.
Summary of the invention
In view of this, can effectively be promoted the purpose of the present invention is to provide a kind of object detection method, apparatus and system
The accuracy rate of object detection results.
To achieve the goals above, technical solution used in the embodiment of the present invention is as follows:
In a first aspect, the embodiment of the invention provides a kind of object detection methods, comprising: acquisition includes target object
Image to be detected；Target detection is carried out to described image to be detected by preset at least two anchor points frame, generates the target
At least two initial detecting results of object；Wherein, every kind of anchor point frame correspondence generates a kind of initial detecting as a result, described at least two
Kind anchor point frame includes the first anchor point frame of the local feature setting based on the target object generic and is based on the mesh
Mark second of anchor point frame of the systemic features setting of object generic；At least two initial detectings result is melted
It closes, obtains the final detection result of the target object.
Further, described that target detection is carried out to described image to be detected by preset at least two anchor points frame, it generates
The step of at least two initial detecting result of the target object, comprising: based on the first described anchor point frame to described to be checked
Altimetric image carries out target detection, generates the first initial detecting result of the target object；The first described initial detecting knot
Fruit includes the local detection block of the target object and the first whole body detection block of the target object；The part detection block position
In in the first whole body detection block；Target detection is carried out to described image to be detected based on second of anchor point frame, is generated
Second of initial detecting result of the target object；Second of initial detecting result includes the second of the target object
Whole body detection block.
Further, described to merge at least two initial detectings result, obtain the final of the target object
The step of testing result, comprising: the first whole body detection block of the target object and the second whole body detection block are merged, institute is obtained
State the final whole body detection block of target object.
Further, described to merge the first whole body detection block of the target object and the second whole body detection block, obtain institute
The step of stating the final whole body detection block of target object, comprising: based on friendship and than algorithm to the first whole body of the target object
Detection block and the second whole body detection block merge processing, obtain the final whole body detection block of the target object.
Further, described that the first whole body detection block of the target object and the second whole body are detected based on friendship and than algorithm
The step of frame merges processing, obtains the final whole body detection block of the target object, comprising: based on the target object
Cutting line is arranged in the position of first whole body detection block；Wherein, the cutting line is by the first whole body detection block of the target object
It is cut into the first upper area and the first lower area；The area of first upper area is not more than first lower region
Domain, and first upper area includes the local detection block；The cutting line examines the second whole body of the target object
It surveys frame and is cut into the second upper area and the second lower area；Calculate first upper area and second upper area
Hand over simultaneously ratio；Based on the friendship and ratio closes the first whole body detection block of the target object and the second whole body detection block
And handle, obtain the final whole body detection block of the target object.
Further, the quantity of the target object is one；It is described based on it is described hand over and ratio to the target object
First whole body detection block and the second whole body detection block merge processing, obtain the final whole body detection block of the target object
Step, comprising: if simultaneously ratio is greater than preset friendship and than threshold value for the friendship, the first whole body detection block is deleted, by institute
State final whole body detection block of the second whole body detection block as the target object.
Further, the quantity of the target object is multiple；It is described to calculate on first upper area and described second
The step of friendship in portion region and ratio, comprising: calculate multiple target objects the first upper area and multiple targets
The friendship of second upper area of object between any two and ratio；It is described based on it is described hand over and ratio to the first of the target object
Whole body detection block and the second whole body detection block merge processing, obtain the step of the final whole body detection block of the target object
Suddenly, comprising: based on the friendship that is calculated and ratio and preset friendship and than threshold value, determine and belong to the of the same target object
One upper area and the second upper area；Wherein, the first upper area and the second upper zone of the same target object are belonged to
Simultaneously friendship of the ratio greater than the first upper area and the second upper area for belonging to different target object and ratio are handed in domain, and return
Belong to the friendship of the first upper area and the second upper area of the same target object and ratio is greater than the friendship and compares threshold value；It is right
In the same target object, the first whole body detection block belonging to the first upper area by the target object is deleted, by the target
Final whole body detection block of the second whole body detection block belonging to second upper area of object as the target object.
Further, the quantity of the target object at least one；The quantity of the first whole body detection block and described second
The quantity of whole body detection block is multiple；It is described based on hand over and than algorithm to the first whole body detection block of the target object and
Before the step of second whole body detection block merges processing, the method also includes: to multiple first whole body detection blocks
Confidence threshold value filtering is carried out with multiple second whole body detection blocks, after obtaining filtered first whole body detection block and filtering
The second whole body detection block；Using non-maxima suppression algorithm to the filtered first whole body detection block and the filtering after
The second whole body detection block handled, obtain the corresponding first whole body detection block of each target object and one
Two whole body detection blocks.
Further, described to multiple first whole body detection blocks and multiple second whole body detection blocks carry out confidence level
Threshold filtering, the step of obtaining filtered first whole body detection block and filtered second whole body detection block, comprising: sentence respectively
Whether the confidence level for each first whole body detection block of breaking is lower than default first confidence threshold value；Wherein, first confidence
Spend the confidence threshold value that threshold value is the local feature setting based on the target object generic；Each described is judged respectively
Whether the confidence level of two whole body detection blocks is lower than default second confidence threshold value；Wherein, second confidence threshold value be based on
The confidence threshold value of the systemic features setting of the target object generic；It will be less than the first of first confidence threshold value
It whole body detection block and is filtered out lower than the second whole body detection block of second confidence threshold value, obtains filtered first whole body inspection
Survey frame and filtered second whole body detection block.
Second aspect, the embodiment of the present invention also provide a kind of object detecting device, comprising: image collection module, for obtaining
Take include target object image to be detected；Module of target detection, for passing through preset at least two anchor points frame to described
Image to be detected carries out target detection, generates at least two initial detecting results of the target object；Wherein, every kind of anchor point frame
Correspondence generates a kind of initial detecting as a result, at least two anchor points frame includes the part based on the target object generic
Second of anchor point frame of the first anchor point frame of feature setting and the systemic features setting based on the target object generic；
Fusion Module obtains the final detection knot of the target object for merging at least two initial detectings result
Fruit.
The third aspect, the embodiment of the invention provides a kind of object detection system, the system comprises: image collector
It sets, processor and storage device；Described image acquisition device, for acquiring image to be detected；It is stored on the storage device
Computer program, the computer program execute such as the described in any item methods of first aspect when being run by the processor.
Fourth aspect, the embodiment of the invention provides a kind of computer readable storage medium, the computer-readable storage
Computer program is stored on medium, the computer program is executed when being run by processor described in above-mentioned any one of first aspect
Method the step of.
The embodiment of the invention provides a kind of object detection methods, apparatus and system, can pass through preset at least two
Anchor point frame carries out target detection to image to be detected of acquisition, generates at least two initial detectings of target object as a result, then
At least two initial detecting results are merged, the final detection result of target object is obtained；Wherein, every kind of anchor point frame is corresponding
A kind of initial detecting is generated as a result, at least two anchor point frames include the of the local feature setting based on target object generic
A kind of second of anchor point frame of anchor point frame and the systemic features setting based on target object generic.The present embodiment can be based on
At least two anchor point frames of different characteristic (the local feature and systemic features) setting of target object generic are to figure to be detected
As carrying out target detection, different initial detectings is obtained as a result, and being merged to obtain by different initial detecting results final
Testing result, be effectively relieved causes feature to lack given object detection results bring shadow because target object is at least partially obscured
It rings, improves the accuracy rate of object detection results.
Other feature and advantage of the embodiment of the present invention will illustrate in the following description, alternatively, Partial Feature and excellent
Point can deduce from specification or unambiguously determine, or the above-mentioned technology by implementing the embodiment of the present invention can obtain
Know.
To enable the above objects, features and advantages of the present invention to be clearer and more comprehensible, preferred embodiment is cited below particularly, and cooperate
Appended attached drawing, is described in detail below.
Detailed description of the invention
It, below will be to specific in order to illustrate more clearly of the specific embodiment of the invention or technical solution in the prior art
Embodiment or attached drawing needed to be used in the description of the prior art be briefly described, it should be apparent that, it is described below
Attached drawing is some embodiments of the present invention, for those of ordinary skill in the art, before not making the creative labor
It puts, is also possible to obtain other drawings based on these drawings.
Fig. 1 shows the structural schematic diagram of a kind of electronic equipment provided by the embodiment of the present invention；
Fig. 2 shows a kind of flow charts of object detection method provided by the embodiment of the present invention；
Fig. 3 shows the structural schematic diagram of the detection network provided by the embodiment of the present invention based on double anchor point frames；
Fig. 4 shows pedestrian detection frame schematic diagram provided by the embodiment of the present invention；
Fig. 5 shows a kind of structural block diagram of object detecting device provided by the embodiment of the present invention.
Specific embodiment
In order to make the object, technical scheme and advantages of the embodiment of the invention clearer, below in conjunction with attached drawing to the present invention
Technical solution be described, it is clear that described embodiments are some of the embodiments of the present invention, rather than whole implementation
Example.In view of causing the accuracy rate of object detection results not high due to target to be detected is blocked in existing target detection technique
The problem of, to improve this problem, the embodiment of the invention provides a kind of object detection method, apparatus and system, which can be answered
For the multiple fields such as automatic Pilot, security protection, new retail, virtual augmented reality, such as detection to personage in market and to road
The detection of road vehicle.It describes in detail below to the embodiment of the present invention.
Embodiment one:
Firstly, describing a kind of object detection method for realizing the embodiment of the present invention, apparatus and system referring to Fig.1
Exemplary electronic device 100.
The structural schematic diagram of a kind of electronic equipment as shown in Figure 1, electronic equipment 100 include one or more processors
102, one or more storage devices 104, input unit 106, output device 108 and image collecting device 110, these components
It is interconnected by bindiny mechanism's (not shown) of bus system 112 and/or other forms.It should be noted that electronic equipment shown in FIG. 1
100 component and structure be it is illustrative, and not restrictive, as needed, the electronic equipment also can have other
Component and structure.
The processor 102 can use digital signal processor (DSP), field programmable gate array (FPGA), can compile
At least one of journey logic array (PLA) example, in hardware realizes that the processor 102 can be central processing unit
(CPU), the processing of graphics processing unit (GPU) or the other forms with data-handling capacity and/or instruction execution capability
The combination of one or more of unit, and the other components that can control in the electronic equipment 100 are desired to execute
Function.
The storage device 104 may include one or more computer program products, and the computer program product can
To include various forms of computer readable storage mediums, such as volatile memory and/or nonvolatile memory.It is described easy
The property lost memory for example may include random access memory (RAM) and/or cache memory (cache) etc..It is described non-
Volatile memory for example may include read-only memory (ROM), hard disk, flash memory etc..In the computer readable storage medium
On can store one or more computer program instructions, processor 102 can run described program instruction, to realize hereafter institute
The client functionality (realized by processor) in the embodiment of the present invention stated and/or other desired functions.In the meter
Can also store various application programs and various data in calculation machine readable storage medium storing program for executing, for example, the application program use and/or
The various data etc. generated.
The input unit 106 can be the device that user is used to input instruction, and may include keyboard, mouse, wheat
One or more of gram wind and touch screen etc..
The output device 108 can export various information (for example, image or sound) to external (for example, user), and
It and may include one or more of display, loudspeaker etc..
Described image acquisition device 110 can shoot the desired image of user (such as photo, video etc.), and will be clapped
The image taken the photograph is stored in the storage device 104 for the use of other components.
Illustratively, for realizing object detection method according to an embodiment of the present invention, the exemplary electron of apparatus and system
Equipment may be implemented as such as first-class intelligent terminal of smart phone, tablet computer, computer, intelligent camera.
Embodiment two:
A kind of object detection method flow chart shown in Figure 2, this method can be set by the electronics that previous embodiment provides
Standby to execute, this method mainly includes the following steps S202~step S206:
Step S202, acquisition include image to be detected of target object.In practical applications, camera etc. can be schemed
As acquisition device shoot include target object original image as image to be detected, be also possible to by network download,
Ground stores or what is manually uploaded includes the image of target object as image to be detected.The image to be detected may include at least
One target object, target object can be people or vehicle etc..
Step S204 carries out target detection to image to be detected by preset at least two anchor points frame, generates target pair
At least two initial detecting results of elephant.Wherein, anchor point frame (anchor box) specifically can be regarded as by area (scale) and
The bounding box that length-width ratio (aspects) the two parameters indicate, the parameter of anchor point frame can be arranged according to actually detected demand,
The different anchor point frame of parameter can be considered different types of anchor point frame.At least two anchor point frames can be based on by detection network to treat
Detection image carries out target detection, and every kind of anchor point frame correspondence generates a kind of initial detecting result；Initial detecting result may include
The position for the target object that image to be detected is included.Above-mentioned at least two anchor points frame includes based on target object generic
Second of anchor point frame of the first anchor point frame of local feature setting and the systemic features setting based on target object generic.
For ease of understanding, it is illustrated so that target object generic is respectively pedestrian and vehicle as an example.If target pair
As generic is pedestrian, the local feature of pedestrian can be face characteristic, the first anchor point frame based on face characteristic setting
It can be using the anchor point frame put and be arranged on the basis of preset certain face's key points；The systemic features of pedestrian include pedestrian from
For head to the global feature of foot, second of anchor point frame of the systemic features setting based on pedestrian, which can be, uses preset certain whole bodies
The anchor point frame put and be arranged on the basis of key point.If target object generic is vehicle, the local feature of vehicle can be
Vehicle license plate characteristic, the first anchor point frame based on vehicle license plate characteristic setting can be using such as license plate central point, license plate corner points as base
The anchor point frame being arranged on schedule；The systemic features of vehicle include the feature of vehicle entirety, the systemic features setting based on vehicle
Second of anchor point frame can be the anchor point frame put on the basis of the key point using such as preset vehicle main portions and be arranged.
At least two initial detecting results are merged, obtain the final detection result of target object by step S206.
Initial detecting result may include corresponding the first the initial detecting result generated of the first anchor point frame and second
Corresponding second of the initial detecting result generated of anchor point frame.In reality scene, often there is body quilt in target object
The case where blocking, heel row human body only has face's exposing such as in group photo and body is blocked, and road vehicle is by trees and extensively
Accuse the shield portions vehicle bodies such as board.Further, since two kinds of anchor point frames are arranged based on different features, and therefore, different initial inspections
The accuracy for surveying result is discrepant, such as referring to above-mentioned by taking pedestrian as an example two kinds of anchor point frames, and corresponding the first is initial
Testing result can detection accuracy to face it is higher, corresponding second of initial detecting result can detection to pedestrian's whole body it is quasi-
True property is higher, but once has the body of pedestrian to be blocked, then second of initial detecting result possibly can not detect trip
People.In turn, the present embodiment merges the first initial detecting result and second of initial detecting result, available target
The more accurate final detection result of object.
Above-mentioned object detection method provided in an embodiment of the present invention, can be based on the different characteristic of target object generic
At least two anchor point frames of (local feature and systemic features) setting carry out target detection to image to be detected, obtain it is different just
Beginning testing result, and merged different initial detecting results to obtain final detection result, it has been effectively relieved because of target pair
Causing feature to lack given object detection results bring as being at least partially obscured influences, and improves the accurate of object detection results
Rate.
In practical applications, present embodiments providing a kind of as shown in Figure 3 detection network based on double anchor point frames (and can
Referred to as detection model) it realizes above by preset at least two anchor points frame to image to be detected progress target detection, generate mesh
The step of marking at least two initial detecting result of object.The detection network based on double anchor point frames is using pedestrian as target pair
As the network detected.As shown in figure 3, the detection network based on double anchor point frames includes backbone network and connects with backbone network
The multiple network branches connect.Wherein, backbone network can use feature pyramid network (FPN, Feature Pyramid
Network it) realizes, the quantity of network branches is multiple, and the quantity of network branches and the characteristic pattern of feature pyramid output
Scale type is corresponding.The input of backbone network is image to be detected, and the output of backbone network is the characteristic pattern of image to be detected.
Each network branches include that face returns main branch, pedestrian returns from branch and pedestrian and returns main branch, and face returns the input of main branch
For the characteristic pattern and the first anchor point frame of image to be detected, export as face detection block.It is to be checked that pedestrian, which returns from the input of branch,
The characteristic pattern of altimetric image and the first anchor point frame, it is first pedestrian's detection block that pedestrian, which returns from the output of branch,.It is understood that
Face datection frame and first pedestrian's detection block are all based on what the first anchor point frame returned, thus first pedestrian's detection block with
Face datection frame has corresponding relationship (binding relationship), in fact, first pedestrian's detection block includes Face datection frame.Pedestrian returns master
The input of branch is the characteristic pattern of second anchor point frame and image to be detected, and the output that pedestrian returns main branch is the second pedestrian detection
Frame.In practical applications, even if the quantity of the target object in image to be detected is one, first pedestrian's detection block and the second row
The quantity of people's detection block may generate the different pedestrian of multiple confidence levels for the same target object and examine to be multiple, such as
Survey frame.
In conjunction with the above-mentioned detection network based on double anchor point frames, target detection is carried out to image to be detected, generates target object
The mode of initial detecting result be referred to following steps (1) and (2):
(1) target detection is carried out to image to be detected based on the first anchor point frame, the first for generating target object is initial
Testing result；The first initial detecting result includes the local detection block of target object and the first whole body detection of target object
Frame；Local detection block is located in the first whole body detection block.
Main branch is returned by face, target detection is carried out based on characteristic pattern of the first anchor point frame to image to be detected, obtain
The local detection block of target object, corresponding to the Face datection frame in Fig. 3.It is returned by pedestrian from branch and is based on the first anchor point frame
Target detection is carried out to the characteristic pattern of image to be detected, obtains the first whole body detection block of target object, corresponding to the in Fig. 3
A group traveling together's detection block.In practical applications, pedestrian's label of first pedestrian's detection block can also be set according to Face datection frame.It can
With understanding, if detection certainly exists the corresponding pedestrian of the face there are face, if but detection there are pedestrians, it is different
Surely certain the case where being also possible that pedestrian's error detection, all in the presence of the face of the pedestrian (the only pedestrian's figure viewed from behind such as detected)
Manikin is such as identified as pedestrian, therefore is that pedestrian's label is arranged in first pedestrian's detection block according further to following three kinds of situations:
Situation one, there are first pedestrian's detection blocks to correspond to Face datection frame, can be by the pedestrian of the first pedestrian detection block
Label is set as 1, indicates pedestrian；Situation two, there are first pedestrian's detection blocks not to have corresponding Face datection frame, can by this
Pedestrian's label of a group traveling together's detection block is set as -1, and the pedestrian is ignored in expression；Situation three, there are in first pedestrian's detection block not
Comprising pedestrian, 0 can be set by pedestrian's label of the first pedestrian detection block, indicate no pedestrian.It in practical applications, can be with
The first pedestrian's detection block for being 1 only for pedestrian's label is handled.
(2) target detection is carried out to image to be detected based on second of anchor point frame, second for generating target object is initial
Testing result；Second of initial detecting result includes the second whole body detection block of target object.
Main branch is returned by pedestrian, target detection is carried out based on characteristic pattern of second of anchor point frame to image to be detected, obtain
Second whole body detection block of target object, corresponding to the second pedestrian detection frame in Fig. 3.
After the first initial detecting result and second of initial detecting result for being obtained based on aforesaid way, by least two
The step of initial detecting result is merged, and the final detection result of target object is obtained may include: by the of target object
One whole body detection block and the fusion of the second whole body detection block, obtain the final whole body detection block of target object.
Furthermore, it is contemplated that including at least one target object, the number of the first whole body detection block of generation in image to be detected
When amount and the quantity of the second whole body detection block are multiple, it will cause melting for the first whole body detection block and the second whole body detection block
Close the big problem of difficulty；To improve this problem, the present embodiment is referred to following steps 1 and step 2 and first detects to the first whole body
Frame and the filtering of the second whole body detection block and processing, then again merge the first whole body detection block and the second whole body detection block.
Step 1, confidence threshold value filtering is carried out to multiple first whole body detection blocks and multiple second whole body detection blocks, obtained
Filtered first whole body detection block and filtered second whole body detection block.
In target detection, multiple positions detection block devious, and each inspection can be generated for each target object
It surveys frame and is corresponding with confidence, the detection block of redundancy can be removed based on confidence level, retain the high detection block of confidence level.In
In a kind of specific implementation, firstly, judging whether the confidence level of each first whole body detection block is lower than default first and sets respectively
Confidence threshold；Wherein, the first confidence threshold value is the confidence threshold value of the local feature setting based on target object generic.
Secondly, judging whether the confidence level of each second whole body detection block is lower than default second confidence threshold value respectively；Its
In, the second confidence threshold value is the confidence threshold value of the systemic features setting based on target object generic.
Above-mentioned first confidence threshold value is mainly based upon local feature setting, and above-mentioned second confidence threshold value is mainly base
In systemic features setting, different set-up mode can from different perspectives on whole body detection block is screened, actually answering
In, the first confidence threshold value and the second confidence threshold value difference, for example 0.3 is set by the first confidence threshold value, by second
Confidence threshold value is set as 0.6.
Finally, will be less than the first whole body detection block of the first confidence threshold value and second complete lower than the second confidence threshold value
Body detection block filters out, and obtains filtered first whole body detection block and filtered second whole body detection block.
In practical applications, the first confidence threshold value can be directly based upon to be filtered the first whole body detection block, may be used also
To be that first will be less than the local detection block of the first confidence threshold value to filter out, filtered local detection block is obtained；It then will filtering
The corresponding first whole body detection block of local detection block afterwards is as filtered first whole body detection block.
Step 2, using non-maxima suppression (Non-Maximum Suppression, NMS) algorithm to filtered first
Whole body detection block and filtered second whole body detection block are handled, and corresponding first whole body of each target object is obtained
Detection block and a second whole body detection block.
Non-maxima suppression algorithm cardinal principle is the element that inhibition is not maximum, it is therefore intended that removes the detection of redundancy
Frame retains optimum detection frame.Such as, it about the processing mode to filtered first whole body detection block, can select to set first
The highest first whole body detection block A of reliability；Then the weight between the first whole body detection block A and other first whole body detection blocks is calculated
Folded degree filters out the first whole body detection block that the degree of overlapping between the first whole body detection block A is more than default degree of overlapping threshold value, most
The corresponding first whole body detection block of each target object is obtained eventually.
About the processing mode to filtered second whole body detection block, above-mentioned obtain to filtered first entirely can refer to
The processing mode of body detection block obtains the corresponding second whole body detection block of each target object, herein not reinflated description.
Further, used by being merged the first whole body detection block of target object and the second whole body detection block
It, can be based on handing over and than (Intersection-over-Union, IoU) algorithm to target object in a kind of specific implementation
The first whole body detection block and the second whole body detection block merge processing, obtain the final whole body detection block of target object.It should
Based on friendship and it can refer to following steps (1) to (3) than the merging treatment mode of algorithm:
(1) cutting line is arranged in the position of the first whole body detection block based on target object；Wherein, cutting line is by target object
The first whole body detection block be cut into the first upper area and the first lower area；The area of first upper area is not more than first
Lower area, and the first upper area includes local detection block；Second whole body detection block of target object is cut by cutting line
Second upper area and the second lower area.
The first whole body detection block is that the first anchor point frame based on local feature setting returns, the inspection of the first whole body
Surveying frame portion corresponding with local feature in frame divides recurrence more accurate, therefore, can be based on frame portion corresponding with local feature
Set up separately and sets cutting line.Referring to pedestrian detection frame schematic diagram shown in Fig. 4, solid box indicates the first being arranged based on face characteristic
First pedestrian's detection block that anchor point frame generates, dotted line frame indicate what second of anchor point frame being arranged based on pedestrian's systemic features was generated
Second pedestrian detection frame.Wherein, the top half of first pedestrian's detection block is more accurate to the recurrence of face, and lower half portion then may be used
Too long or too short situation can occur, such as first pedestrian's detection block is not covered with the knee of pedestrian with lower portion；In order to make to examine
The cutting result for surveying frame is more accurate, can set cutting line to the top half of first pedestrian's detection block, such as along first
The middle line that pedestrian detection frame Vertical Square is set up.First whole body detection block of target object is cut on first by the cutting line
Portion region and the first lower area, and the second whole body detection block of target object is cut under the second upper area and second
Portion region.
(2) friendship and the ratio of the first upper area and the second upper area are calculated.
In the present embodiment, if the quantity of target object is one, the first upper area and second can directly be calculated
The friendship of upper area and ratio.If the quantity of target object be it is multiple, the first upper zone of multiple target objects can be calculated
The friendship of second upper area of domain and multiple target objects between any two and ratio.It hands over and ratio is the first upper area and second
The intersection of the area of upper area and the ratio of union.It is illustrated as follows in order to make it easy to understand, providing: the first upper area packet
A1, b1, c1 are included, the second upper area includes a2, b2, d2, then the friendship and ratio for calculating separately a1 and a2 (are represented by
IOUa1-a2), the friendship of a1 and b2 and ratio I OUa1-b2, a1 and d2 friendship and ratio I OUa1-d2, b1 and a2 friendship and ratio
IOUb1-a2, b1 and b2 friendship and ratio I OUb1-b2, b1 and d2 friendship and ratio I OUb1-d2, c1 and a2 friendship and ratio I OUc1-a2、
The friendship of c1 and b2 and ratio I OUc1-b2, c1 and d2 friendship and ratio I OUc1-d2。
Simultaneously ratio can characterize the first whole body detection block and the second whole body for the friendship of first upper area and the second upper area
The degree of overlapping of detection block only hand over and than counting to the upper half area that two whole body detection blocks are divided through same cutting line
It calculates, not only accuracy rate is higher, but also can significantly reduce calculation amount, improves treatment effeciency.
(3) based on friendship, simultaneously ratio merges place to the first whole body detection block of target object and the second whole body detection block
Reason, obtains the final whole body detection block of target object.
In view of the first anchor point frame is arranged based on local feature, the corresponding position for obtaining local detection block is returned
Return comparison accurate, but the accuracy that the position of corresponding the first obtained whole body detection block returns is slightly lower.Second of anchor point frame is base
It is arranged in systemic features, it is more accurate that the position of corresponding the second obtained whole body detection block returns.Therefore, by first
When whole body detection block and the second whole body detection block merge processing, it can be replaced using the higher second whole body detection block of accuracy
Change the first slightly lower whole body detection block of accuracy.Quantity below for target object is one or more scenes respectively to inspection
Description is unfolded in the merging treatment process for surveying frame.
Quantity for target object is one scene: if handed over and ratio is greater than preset friendship and than threshold value, by the
One whole body detection block is deleted, using the second whole body detection block as the final whole body detection block of target object.
For target object quantity be multiple scenes: the merging treatment process of detection block may include following steps a and
Step b:
Step a, based on the friendship and ratio and preset friendship that are calculated and than threshold value, determination belongs to the same target pair
The first upper area and the second upper area of elephant.Wherein, the first upper area and second of the same target object is belonged to
Upper area hands over simultaneously friendship of the ratio greater than the first upper area and the second upper area for belonging to different target object and ratio
Value, and belong to the first upper area and the second upper area of the same target object friendship and ratio be greater than hand over and compare threshold
Value.
In practical applications, will hand over and ratio be greater than preset friendship and than the first upper area of threshold value (such as 0.3) and
Second upper area is determined as candidate first upper area and candidate second upper area；Calculate multiple candidate first upper areas
With the friendship of multiple candidate second upper areas between any two and ratio, determined on each candidate first based on the friendship and ratio of calculating
Corresponding the second upper area of best candidate in portion region, second upper area of best candidate and candidate's the first upper area weight
Folded degree highest.For ease of understanding, it is illustrated below: being computed the friendship for determining the second upper area b2 and the first upper area a1
And ratio is maximum, then the second upper area b2 is the second upper area of best candidate of the first upper area a1, similarly, it is assumed that the
Two upper area b1 are corresponding optimal second upper area of the first upper area a2.And determine each candidate second upper zone
Corresponding the first upper area of best candidate in domain, such as: it is computed and determines the first upper area a1's and the second upper area b2
It hands over and ratio is maximum, then the first upper area a1 is the first upper area of best candidate of the second upper area b2, but is computed
Determine that the first upper area of best candidate that the second upper area is b1 is c1.Then by the first upper area a1 and the second upper zone
Domain b2 is determined as two upper areas of optimal object each other, and the first upper area a1 and the second upper area b2 belong to same
Target object.
Step b, for the same target object, the detection of the first whole body belonging to the first upper area by the target object
Frame is deleted, using the second whole body detection block belonging to the second upper area of the target object as the final whole body of the target object
Detection block.
It is understood that the first whole body detection block generated based on the first anchor point frame and local detection block with based on the
The second whole body detection block that two kinds of anchor point frames generate, in addition to including above-mentioned belonging to the same target object and can be with merging treatment
Detection block except, it is also possible to the whole body detection block including other target objects can will belong to same in practical applications
The first whole body detection block, the second whole body detection block of whole body detection block and other target objects after the merging of a target object are total
With testing result is used as, the problem low because of target object recall rate caused by blocking can be alleviated, can effectively promote target
The detection accuracy of object；Wherein, which can be understood as real in destination number and image to be detected in testing result
The ratio of the included destination number in border can be used for preferably measuring object detection method to the recall ratio of target to be detected.Such as,
When including 10 pedestrians in image to be detected, there is the lower half body of 2 pedestrians to be blocked and only expose above the waist, it is existing
Object detection method be typically only capable to detect pedestrian's whole body, therefore be only capable of detecting 8 pedestrians and can not being tested with
2 pedestrians that body is blocked, then corresponding recall rate is 80%；And above-mentioned object detection method provided in this embodiment, no
But it is capable of detecting when be not blocked 8 pedestrians, additionally it is possible to be tested with 2 pedestrians that body is blocked, then it is corresponding to recall
Rate is 100%, effectively improves the accuracy and reliability of target detection.
To sum up, above-described embodiment being capable of the different characteristic (local feature and systemic features) based on target object generic
At least two anchor point frames being arranged carry out target detection to image to be detected, obtain different initial detectings as a result, and will be different
Initial detecting result merged to obtain final detection result, being effectively relieved leads to spy because target object is at least partially obscured
Sign, which lacks given object detection results bring, to be influenced, and the accuracy rate of object detection results is improved.
Embodiment three:
For object detection method provided in embodiment two, the embodiment of the invention provides a kind of target detection dresses
It sets, a kind of structural block diagram of object detecting device shown in Figure 5, comprising:
Image collection module 502, for obtain include target object image to be detected；
Module of target detection 504, for carrying out target detection to image to be detected by preset at least two anchor points frame,
Generate at least two initial detecting results of target object；Wherein, every kind of anchor point frame correspondence generates a kind of initial detecting as a result, extremely
Few two kinds of anchor point frames include the first anchor point frame of the local feature setting based on target object generic and are based on target pair
Second of anchor point frame being arranged as the systemic features of generic；
Fusion Module 506 obtains the final detection of target object for merging at least two initial detecting results
As a result.
Object detecting device provided in an embodiment of the present invention, can by preset at least two anchor points frame to acquisition to
Detection image carries out target detection, generates at least two initial detectings of target object as a result, then by least two initial inspections
It surveys result to be merged, obtains the final detection result of target object；Wherein, every kind of anchor point frame correspondence generates a kind of initial detecting
As a result, at least two anchor point frames include the first anchor point frame of the local feature setting based on target object generic and are based on
Second of anchor point frame of the systemic features setting of target object generic.The present embodiment can be based on target object generic
Different characteristic (local feature and systemic features) setting at least two anchor point frames to image to be detected carry out target detection, obtain
To different initial detectings as a result, and merged different initial detecting results to obtain final detection result, be effectively relieved
Cause feature to lack given object detection results bring influence because target object is at least partially obscured, improves target detection
As a result accuracy rate.
In one embodiment, above-mentioned module of target detection 504 is also used to: based on the first anchor point frame to figure to be detected
As carrying out target detection, the first initial detecting result of target object is generated；The first initial detecting result includes target pair
The local detection block of elephant and the first whole body detection block of target object；Local detection block is located in the first whole body detection block；It is based on
Second of anchor point frame carries out target detection to image to be detected, generates second of initial detecting result of target object；Second
Initial detecting result includes the second whole body detection block of target object.
In one embodiment, above-mentioned Fusion Module 506 is also used to: by the first whole body detection block of target object and
The fusion of two whole body detection blocks, obtains the final whole body detection block of target object.
In one embodiment, above-mentioned Fusion Module 506 is also used to: based on hand over and than algorithm to target object first
Whole body detection block and the second whole body detection block merge processing, obtain the final whole body detection block of target object.
In one embodiment, above-mentioned Fusion Module 506 specifically includes: cutting line setting unit, for being based on target
Cutting line is arranged in the position of first whole body detection block of object；Wherein, cutting line cuts the first whole body detection block of target object
It is segmented into the first upper area and the first lower area；The area of first upper area is not more than the first lower area, and on first
Portion region includes local detection block；Second whole body detection block of target object is cut into the second upper area and second by cutting line
Lower area；Simultaneously ratio calculation unit is handed over, for calculating friendship and the ratio of the first upper area and the second upper area；At merging
Unit is managed, for simultaneously ratio to merge place to the first whole body detection block of target object and the second whole body detection block based on friendship
Reason, obtains the final whole body detection block of target object.
In one embodiment, the quantity of target object is one；Above-mentioned Fusion Module 506 is also used to: if handed over simultaneously
Ratio is greater than preset friendship and than threshold value, the first whole body detection block is deleted, using the second whole body detection block as target object
Final whole body detection block.
In one embodiment, the quantity of target object is multiple；Simultaneously ratio calculation unit is also used to for above-mentioned friendship: being calculated
The friendship between any two of second upper area of the first upper area of multiple target objects and multiple target objects and ratio；It is above-mentioned
Merge processing unit is also used to: based on the friendship and ratio and preset friendship that are calculated and than threshold value, determination belongs to same
The first upper area and the second upper area of target object；Wherein, the first upper area of the same target object is belonged to
Friendship and ratio with the second upper area are greater than the first upper area and the second upper area for belonging to different target object
Hand over and ratio, and belong to the same target object the first upper area and the second upper area friendship and ratio be greater than hand over simultaneously
Compare threshold value；For the same target object, the first whole body detection block belonging to the first upper area by the target object is deleted,
Using the second whole body detection block belonging to the second upper area of the target object as the final whole body detection block of the target object.
In one embodiment, the quantity of target object is at least one；The quantity and second of first whole body detection block
The quantity of whole body detection block is multiple；Above-mentioned object detecting device further includes filtering module (not shown)；The filter module
Block is used for: confidence threshold value filtering is carried out to multiple first whole body detection blocks and multiple second whole body detection blocks, after obtaining filtering
The first whole body detection block and filtered second whole body detection block；It is complete to filtered first using non-maxima suppression algorithm
Body detection block and filtered second whole body detection block are handled, and corresponding first whole body inspection of each target object is obtained
Survey frame and a second whole body detection block.
In one embodiment, above-mentioned filtering module is also used to: judging the confidence of each first whole body detection block respectively
Whether degree is lower than default first confidence threshold value；Wherein, the first confidence threshold value is the part based on target object generic
The confidence threshold value of feature setting；Judge whether the confidence level of each second whole body detection block is lower than default second confidence level respectively
Threshold value；Wherein, the second confidence threshold value is the confidence threshold value of the systemic features setting based on target object generic；It will be low
The first whole body detection block in the first confidence threshold value and the second whole body detection block lower than the second confidence threshold value filter out, and obtain
Filtered first whole body detection block and filtered second whole body detection block.
The technical effect and previous embodiment two of device provided by the present embodiment, realization principle and generation are identical, are
It briefly describes, Installation practice part does not refer to place, can refer to corresponding contents in preceding method embodiment.
Example IV:
For object detection method provided in embodiment two, the embodiment of the invention provides a kind of target detection systems
System, the system comprises: image collecting device, processor and storage device；Described image acquisition device, it is to be detected for acquiring
Image；Computer program is stored on the storage device, the computer program executes such as when being run by the processor
The object detection method of above-described embodiment two.
It is apparent to those skilled in the art that for convenience and simplicity of description, the system of foregoing description
Specific work process can refer to corresponding processes in the foregoing method embodiment, and details are not described herein.
Further, the present embodiment additionally provides a kind of computer readable storage medium, the computer readable storage medium
On be stored with computer program, the computer program is executed when being run by processor provided by above-mentioned preceding method embodiment
The step of object detection method.
The computer program product of object detection method, apparatus and system provided by the embodiment of the present invention, including storage
The computer readable storage medium of program code, the instruction that said program code includes can be used for executing previous methods embodiment
Described in method, specific implementation can be found in embodiment of the method, details are not described herein.
In addition, in the description of the embodiment of the present invention unless specifically defined or limited otherwise, term " installation ", " phase
Even ", " connection " shall be understood in a broad sense, for example, it may be being fixedly connected, may be a detachable connection, or be integrally connected；It can
To be mechanical connection, it is also possible to be electrically connected；It can be directly connected, can also can be indirectly connected through an intermediary
Connection inside two elements.For the ordinary skill in the art, above-mentioned term can be understood at this with concrete condition
Concrete meaning in invention.
It, can be with if the function is realized in the form of SFU software functional unit and when sold or used as an independent product
It is stored in a computer readable storage medium.Based on this understanding, technical solution of the present invention is substantially in other words
The part of the part that contributes to existing technology or the technical solution can be embodied in the form of software products, the meter
Calculation machine software product is stored in a storage medium, including some instructions are used so that a computer equipment (can be a
People's computer, server or network equipment etc.) it performs all or part of the steps of the method described in the various embodiments of the present invention.
And storage medium above-mentioned includes: that USB flash disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), arbitrary access are deposited
The various media that can store program code such as reservoir (RAM, Random Access Memory), magnetic or disk.
In the description of the present invention, it should be noted that term " center ", "upper", "lower", "left", "right", "vertical",
The orientation or positional relationship of the instructions such as "horizontal", "inner", "outside" be based on the orientation or positional relationship shown in the drawings, merely to
Convenient for description the present invention and simplify description, rather than the device or element of indication or suggestion meaning must have a particular orientation,
It is constructed and operated in a specific orientation, therefore is not considered as limiting the invention.In addition, term " first ", " second ",
" third " is used for descriptive purposes only and cannot be understood as indicating or suggesting relative importance.
Finally, it should be noted that embodiment described above, only a specific embodiment of the invention, to illustrate the present invention
Technical solution, rather than its limitations, scope of protection of the present invention is not limited thereto, although with reference to the foregoing embodiments to this hair
It is bright to be described in detail, those skilled in the art should understand that: anyone skilled in the art
In the technical scope disclosed by the present invention, it can still modify to technical solution documented by previous embodiment or can be light
It is readily conceivable that variation or equivalent replacement of some of the technical features；And these modifications, variation or replacement, do not make
The essence of corresponding technical solution is detached from the spirit and scope of technical solution of the embodiment of the present invention, should all cover in protection of the invention
Within the scope of.Therefore, protection scope of the present invention should be based on the protection scope of the described claims.
Claims (12)
1. a kind of object detection method characterized by comprising
Acquisition includes image to be detected of target object；
Target detection is carried out to described image to be detected by preset at least two anchor points frame, generates the target object extremely
Few two kinds of initial detecting results；Wherein, every kind of anchor point frame correspondence generates a kind of initial detecting as a result, at least two anchor points frame
Including based on the target object generic local feature setting the first anchor point frame and be based on the target object institute
Belong to second of anchor point frame of the systemic features setting of classification；
At least two initial detectings result is merged, the final detection result of the target object is obtained.
2. the method according to claim 1, wherein it is described by preset at least two anchor points frame to it is described to
The step of detection image carries out target detection, generates at least two initial detecting result of the target object, comprising:
Target detection is carried out to described image to be detected based on the first described anchor point frame, generates the first of the target object
Initial detecting result；The first described initial detecting result include the target object local detection block and the target object
The first whole body detection block；The part detection block is located in the first whole body detection block；
Target detection is carried out to described image to be detected based on second of anchor point frame, generates second of the target object
Initial detecting result；Second of initial detecting result includes the second whole body detection block of the target object.
3. according to the method described in claim 2, it is characterized in that, described melt at least two initial detectings result
The step of closing, obtaining the final detection result of the target object, comprising:
First whole body detection block of the target object and the second whole body detection block are merged, the final of the target object is obtained
Whole body detection block.
4. according to the method described in claim 3, it is characterized in that, the first whole body detection block by the target object and
The step of second whole body detection block merges, and obtains the final whole body detection block of the target object, comprising:
Based on friendship and processing is merged to the first whole body detection block of the target object and the second whole body detection block than algorithm,
Obtain the final whole body detection block of the target object.
5. according to the method described in claim 4, it is characterized in that, it is described based on hand over and than algorithm to the target object the
One whole body detection block and the second whole body detection block merge processing, obtain the step of the final whole body detection block of the target object
Suddenly, comprising:
Cutting line is arranged in the position of the first whole body detection block based on the target object；Wherein, the cutting line is by the mesh
First whole body detection block of mark object is cut into the first upper area and the first lower area；The area of first upper area
No more than first lower area, and first upper area includes the local detection block；The cutting line will be described
Second whole body detection block of target object is cut into the second upper area and the second lower area；
Calculate the friendship of first upper area and second upper area and ratio；
Based on the friendship and ratio merges place to the first whole body detection block of the target object and the second whole body detection block
Reason, obtains the final whole body detection block of the target object.
6. according to the method described in claim 5, it is characterized in that, the quantity of the target object is one；
It is described that based on the friendship, simultaneously ratio closes the first whole body detection block of the target object and the second whole body detection block
And the step of handling, obtaining the final whole body detection block of the target object, comprising:
If the friendship and ratio are greater than preset friendship and than threshold value, the first whole body detection block is deleted, by described second
Final whole body detection block of the whole body detection block as the target object.
7. according to the method described in claim 5, it is characterized in that, the quantity of the target object is multiple；
The step of friendship for calculating first upper area and second upper area and ratio, comprising:
Second upper area of the first upper area and multiple target objects that calculate multiple target objects two-by-two it
Between friendship and ratio；
It is described that based on the friendship, simultaneously ratio closes the first whole body detection block of the target object and the second whole body detection block
And the step of handling, obtaining the final whole body detection block of the target object, comprising:
Based on the friendship and ratio and preset friendship that are calculated and than threshold value, determination is belonged on the first of the same target object
Portion region and the second upper area；Wherein, the first upper area and second upper area of the same target object are belonged to
It hands over and ratio is greater than the friendship for belonging to the first upper area and the second upper area of different target object and ratio, and belong to
Simultaneously ratio is greater than the friendship and compares threshold value for the friendship of the first upper area and the second upper area of the same target object；
For the same target object, the first whole body detection block belonging to the first upper area by the target object is deleted, will
Final whole body detection block of the second whole body detection block belonging to second upper area of the target object as the target object.
8. according to the method described in claim 4, it is characterized in that, the quantity of the target object is at least one；Described
The quantity of the quantity of one whole body detection block and the second whole body detection block is multiple；
It is described based on hand over and the first whole body detection block of the target object and the second whole body detection block are closed than algorithm
And before the step of handling, the method also includes:
Confidence threshold value filtering is carried out to multiple first whole body detection blocks and multiple second whole body detection blocks, is obtained
The first whole body detection block and filtered second whole body detection block after filter；
The filtered first whole body detection block and filtered second whole body are examined using non-maxima suppression algorithm
It surveys frame to be handled, obtains the corresponding first whole body detection block of each target object and the second whole body detection
Frame.
9. according to the method described in claim 8, it is characterized in that, described to multiple first whole body detection blocks and multiple institutes
It states the second whole body detection block and carries out confidence threshold value filtering, obtain filtered first whole body detection block and filtered second entirely
The step of body detection block, comprising:
Judge whether the confidence level of each first whole body detection block is lower than default first confidence threshold value respectively；Wherein, institute
State the confidence threshold value that the first confidence threshold value is the local feature setting based on the target object generic；
Judge whether the confidence level of each second whole body detection block is lower than default second confidence threshold value respectively；Wherein, institute
State the confidence threshold value that the second confidence threshold value is the systemic features setting based on the target object generic；
It will be less than the first whole body detection block of first confidence threshold value and second complete lower than second confidence threshold value
Body detection block filters out, and obtains filtered first whole body detection block and filtered second whole body detection block.
10. a kind of object detecting device characterized by comprising
Image collection module, for obtain include target object image to be detected；
Module of target detection, it is raw for carrying out target detection to described image to be detected by preset at least two anchor points frame
At at least two initial detecting results of the target object；Wherein, every kind of anchor point frame it is corresponding generate a kind of initial detecting as a result,
At least two anchor points frame include based on the target object generic local feature setting the first anchor point frame and
Second of anchor point frame of the systemic features setting based on the target object generic；
Fusion Module obtains the most final inspection of the target object for merging at least two initial detectings result
Survey result.
11. a kind of object detection system, which is characterized in that the system comprises: image collecting device, processor and storage dress
It sets；
Described image acquisition device, for acquiring image to be detected；
Computer program is stored on the storage device, the computer program is executed when being run by the processor as weighed
Benefit requires 1 to 9 described in any item methods.
12. a kind of computer readable storage medium, computer program, feature are stored on the computer readable storage medium
The step of being, the described in any item methods of the claims 1 to 9 executed when the computer program is run by processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910830745.3A CN110532985B (en) | 2019-09-02 | 2019-09-02 | Target detection method, device and system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910830745.3A CN110532985B (en) | 2019-09-02 | 2019-09-02 | Target detection method, device and system |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN110532985A true CN110532985A (en) | 2019-12-03 |
| CN110532985B CN110532985B (en) | 2022-07-22 |
Family
ID=68666621
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910830745.3A Active CN110532985B (en) | 2019-09-02 | 2019-09-02 | Target detection method, device and system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN110532985B (en) |
Cited By (15)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275002A (en) * | 2020-02-18 | 2020-06-12 | 上海商汤临港智能科技有限公司 | Image processing method and device and electronic equipment |
| CN111353473A (en) * | 2020-03-30 | 2020-06-30 | 浙江大华技术股份有限公司 | Face detection method and device, electronic equipment and storage medium |
| CN111507327A (en) * | 2020-04-07 | 2020-08-07 | 浙江大华技术股份有限公司 | Target detection method and device |
| CN111507286A (en) * | 2020-04-22 | 2020-08-07 | 北京爱笔科技有限公司 | Dummy detection method and device |
| CN111582177A (en) * | 2020-05-09 | 2020-08-25 | 北京爱笔科技有限公司 | Image detection method and related device |
| CN111597959A (en) * | 2020-05-12 | 2020-08-28 | 三一重工股份有限公司 | Behavior detection method and device and electronic equipment |
| CN111814612A (en) * | 2020-06-24 | 2020-10-23 | 浙江大华技术股份有限公司 | Target face detection method and related device thereof |
| CN111860493A (en) * | 2020-06-12 | 2020-10-30 | 北京图森智途科技有限公司 | Target detection method and device based on point cloud data |
| CN112257692A (en) * | 2020-12-22 | 2021-01-22 | 湖北亿咖通科技有限公司 | Pedestrian target detection method, electronic device and storage medium |
| CN112287808A (en) * | 2020-10-27 | 2021-01-29 | 江苏云从曦和人工智能有限公司 | Motion trajectory analysis warning method, device, system and storage medium |
| CN112949526A (en) * | 2021-03-12 | 2021-06-11 | 深圳海翼智新科技有限公司 | Face detection method and device |
| CN113537064A (en) * | 2021-07-16 | 2021-10-22 | 河南牧原智能科技有限公司 | Weak pig automatic detection marking method and system |
| CN113592902A (en) * | 2021-06-21 | 2021-11-02 | 北京迈格威科技有限公司 | Target tracking method and device, computer equipment and storage medium |
| CN113743261A (en) * | 2021-08-23 | 2021-12-03 | 河南牧原智能科技有限公司 | Pig body trauma detection method and device and readable storage medium |
| CN114120127A (en) * | 2021-11-30 | 2022-03-01 | 济南博观智能科技有限公司 | Target detection method, device and related equipment |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180025249A1 (en) * | 2016-07-25 | 2018-01-25 | Mitsubishi Electric Research Laboratories, Inc. | Object Detection System and Object Detection Method |
| CN108520229A (en) * | 2018-04-04 | 2018-09-11 | 北京旷视科技有限公司 | Image detecting method, device, electronic equipment and computer-readable medium |
| CN108694401A (en) * | 2018-05-09 | 2018-10-23 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
| CN108898047A (en) * | 2018-04-27 | 2018-11-27 | 中国科学院自动化研究所 | The pedestrian detection method and system of perception are blocked based on piecemeal |
| CN109145898A (en) * | 2018-07-26 | 2019-01-04 | 清华大学深圳研究生院 | A kind of object detecting method based on convolutional neural networks and iterator mechanism |
| CN109766796A (en) * | 2018-12-20 | 2019-05-17 | 西华大学 | A kind of depth pedestrian detection method towards dense population |
| CN110059547A (en) * | 2019-03-08 | 2019-07-26 | 北京旷视科技有限公司 | Object detection method and device |
| CN110096933A (en) * | 2018-01-30 | 2019-08-06 | 华为技术有限公司 | The method, apparatus and system of target detection |
| US20190251333A1 (en) * | 2017-06-02 | 2019-08-15 | Tencent Technology (Shenzhen) Company Limited | Face detection training method and apparatus, and electronic device |
-
2019
- 2019-09-02 CN CN201910830745.3A patent/CN110532985B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180025249A1 (en) * | 2016-07-25 | 2018-01-25 | Mitsubishi Electric Research Laboratories, Inc. | Object Detection System and Object Detection Method |
| US20190251333A1 (en) * | 2017-06-02 | 2019-08-15 | Tencent Technology (Shenzhen) Company Limited | Face detection training method and apparatus, and electronic device |
| CN110096933A (en) * | 2018-01-30 | 2019-08-06 | 华为技术有限公司 | The method, apparatus and system of target detection |
| CN108520229A (en) * | 2018-04-04 | 2018-09-11 | 北京旷视科技有限公司 | Image detecting method, device, electronic equipment and computer-readable medium |
| CN108898047A (en) * | 2018-04-27 | 2018-11-27 | 中国科学院自动化研究所 | The pedestrian detection method and system of perception are blocked based on piecemeal |
| CN108694401A (en) * | 2018-05-09 | 2018-10-23 | 北京旷视科技有限公司 | Object detection method, apparatus and system |
| CN109145898A (en) * | 2018-07-26 | 2019-01-04 | 清华大学深圳研究生院 | A kind of object detecting method based on convolutional neural networks and iterator mechanism |
| CN109766796A (en) * | 2018-12-20 | 2019-05-17 | 西华大学 | A kind of depth pedestrian detection method towards dense population |
| CN110059547A (en) * | 2019-03-08 | 2019-07-26 | 北京旷视科技有限公司 | Object detection method and device |
Cited By (23)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111275002A (en) * | 2020-02-18 | 2020-06-12 | 上海商汤临港智能科技有限公司 | Image processing method and device and electronic equipment |
| CN111353473B (en) * | 2020-03-30 | 2023-04-14 | 浙江大华技术股份有限公司 | Face detection method and device, electronic equipment and storage medium |
| CN111353473A (en) * | 2020-03-30 | 2020-06-30 | 浙江大华技术股份有限公司 | Face detection method and device, electronic equipment and storage medium |
| CN111507327A (en) * | 2020-04-07 | 2020-08-07 | 浙江大华技术股份有限公司 | Target detection method and device |
| CN111507327B (en) * | 2020-04-07 | 2023-04-14 | 浙江大华技术股份有限公司 | Target detection method and device |
| CN111507286A (en) * | 2020-04-22 | 2020-08-07 | 北京爱笔科技有限公司 | Dummy detection method and device |
| CN111507286B (en) * | 2020-04-22 | 2023-05-02 | 北京爱笔科技有限公司 | Dummy detection method and device |
| CN111582177A (en) * | 2020-05-09 | 2020-08-25 | 北京爱笔科技有限公司 | Image detection method and related device |
| CN111597959A (en) * | 2020-05-12 | 2020-08-28 | 三一重工股份有限公司 | Behavior detection method and device and electronic equipment |
| CN111597959B (en) * | 2020-05-12 | 2023-09-26 | 盛景智能科技(嘉兴)有限公司 | Behavior detection method and device and electronic equipment |
| CN111860493A (en) * | 2020-06-12 | 2020-10-30 | 北京图森智途科技有限公司 | Target detection method and device based on point cloud data |
| CN111860493B (en) * | 2020-06-12 | 2024-02-09 | 北京图森智途科技有限公司 | Target detection method and device based on point cloud data |
| CN111814612A (en) * | 2020-06-24 | 2020-10-23 | 浙江大华技术股份有限公司 | Target face detection method and related device thereof |
| CN112287808A (en) * | 2020-10-27 | 2021-01-29 | 江苏云从曦和人工智能有限公司 | Motion trajectory analysis warning method, device, system and storage medium |
| WO2022134624A1 (en) * | 2020-12-22 | 2022-06-30 | 亿咖通(湖北)技术有限公司 | Pedestrian target detection method, electronic device and storage medium |
| CN112257692A (en) * | 2020-12-22 | 2021-01-22 | 湖北亿咖通科技有限公司 | Pedestrian target detection method, electronic device and storage medium |
| CN112949526A (en) * | 2021-03-12 | 2021-06-11 | 深圳海翼智新科技有限公司 | Face detection method and device |
| CN112949526B (en) * | 2021-03-12 | 2024-03-29 | 深圳海翼智新科技有限公司 | Face detection method and device |
| CN113592902A (en) * | 2021-06-21 | 2021-11-02 | 北京迈格威科技有限公司 | Target tracking method and device, computer equipment and storage medium |
| CN113537064A (en) * | 2021-07-16 | 2021-10-22 | 河南牧原智能科技有限公司 | Weak pig automatic detection marking method and system |
| CN113743261A (en) * | 2021-08-23 | 2021-12-03 | 河南牧原智能科技有限公司 | Pig body trauma detection method and device and readable storage medium |
| CN114120127A (en) * | 2021-11-30 | 2022-03-01 | 济南博观智能科技有限公司 | Target detection method, device and related equipment |
| CN114120127B (en) * | 2021-11-30 | 2024-06-07 | 济南博观智能科技有限公司 | Target detection method, device and related equipment |
Also Published As
| Publication number | Publication date |
|---|---|
| CN110532985B (en) | 2022-07-22 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110532985A (en) | Object detection method, apparatus and system | |
| CN106960195A (en) | A kind of people counting method and device based on deep learning | |
| CN106296721B (en) | Object aggregation detection method and device based on stereoscopic vision | |
| Qin et al. | 3D change detection at street level using mobile laser scanning point clouds and terrestrial images | |
| CN105793730B (en) | The classification based on laser radar of object motion | |
| CN108520229A (en) | Image detecting method, device, electronic equipment and computer-readable medium | |
| CN109376667A (en) | Object detection method, device and electronic equipment | |
| Mallikarjuna et al. | Traffic data collection under mixed traffic conditions using video image processing | |
| CN108133495A (en) | For the 3D method for reconstructing, corresponding program product and equipment of mobile device environment | |
| CN107292886A (en) | Object intrusion detection method and device based on mesh generation and neutral net | |
| CN111368600A (en) | Method and device for detecting and identifying remote sensing image target, readable storage medium and equipment | |
| JP2015062121A (en) | System and method of alerting driver that visual perception of pedestrian may be difficult | |
| CN108737793A (en) | Use camera network moving object tracking | |
| CN106971178A (en) | Pedestrian detection and the method and device recognized again | |
| CN107909027A (en) | It is a kind of that there is the quick human body target detection method for blocking processing | |
| CN108875750A (en) | object detecting method, device and system and storage medium | |
| CN110245579A (en) | Density of stream of people prediction technique and device, computer equipment and readable medium | |
| CN106803083A (en) | The method and device of pedestrian detection | |
| KR101840042B1 (en) | Multi-Imaginary Fence Line Setting Method and Trespassing Sensing System | |
| CN105243356A (en) | Method of building pedestrian detection model and device and pedestrian detection method | |
| CN110067274A (en) | Apparatus control method and excavator | |
| WO2020217812A1 (en) | Image processing device that recognizes state of subject and method for same | |
| CN115565207B (en) | Occlusion scene downlink person detection method with feature simulation fused | |
| CN109447325A (en) | Precipitation data detection method, device and electronic equipment based on random forests algorithm | |
| CN107689157A (en) | Traffic intersection based on deep learning can passing road planing method |
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 |