CN109284665A - Method and apparatus for reducing the detection candidate quantity of object identifying method - Google Patents
Method and apparatus for reducing the detection candidate quantity of object identifying method Download PDFInfo
- Publication number
- CN109284665A CN109284665A CN201810789967.0A CN201810789967A CN109284665A CN 109284665 A CN109284665 A CN 109284665A CN 201810789967 A CN201810789967 A CN 201810789967A CN 109284665 A CN109284665 A CN 109284665A
- Authority
- CN
- China
- Prior art keywords
- candidate
- detection candidate
- detection
- list
- score
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
- G06V20/582—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads of traffic signs
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/211—Selection of the most significant subset of features
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/24—Aligning, centring, orientation detection or correction of the image
- G06V10/245—Aligning, centring, orientation detection or correction of the image by locating a pattern; Special marks for positioning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/30—Noise filtering
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/36—Applying a local operator, i.e. means to operate on image points situated in the vicinity of a given point; Non-linear local filtering operations, e.g. median filtering
-
- 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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
- G06V20/58—Recognition of moving objects or obstacles, e.g. vehicles or pedestrians; Recognition of traffic objects, e.g. traffic signs, traffic lights or roads
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Nonlinear Science (AREA)
- Data Mining & Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The present invention relates to a kind of methods for reducing the quantity of the detection candidate (118) of object identifying method, wherein method includes the following steps: by least using score (120) threshold value (126) list from the list (122) of evaluation detection candidate (118) for accordingly distributing to detection candidate (118)；Delete (504) detection candidate (118), the detection candidate (118) for the threshold value that score (120) are less than in threshold value (126) list is wherein deleted from list (122), particularly as long as it is no more than the maximum distance with the detection candidate (118) for calculating threshold value.
Description
Technical field
The present invention relates to a kind of method and apparatus for reducing the detection candidate quantity of object identifying method.The present invention
Content further include a kind of computer program.
Background technique
In order to search the object for there are corner angle in the picture, corner region can be searched in the picture.The corner region phase of object
For arranging in known manner each other.If multiple corner regions are located in specific geometrical boundary condition, may recognize that
The candidate of the object.
Summary of the invention
In this context, it is proposed according to the present invention using the scheme introduced herein a kind of for reducing object identifying method
Detection candidate quantity method, further provide the device using this method, a kind of image processing system, and last
Propose a kind of corresponding computer program.
When searching for corner angle, the small search window in search pattern can be guided on the image.If in the search window
Possible corner region is had found, then can mark the position of search window and is stored as corner region candidate.In addition, can
Following assessment is carried out, i.e., how reliably to identify the feature of corner angle.
This search technique is very fast, but can obtain a large amount of corner region candidates, and many is wrong.
In the scheme introduced herein, abandoned those relative to having the candidate well evaluated in its local environment and
The candidate of lower assessment valence is obtained.Since considered candidate quantity is reduced, then the corner region being mutually matched is being waited
Reduce required amount of calculation in the search of the person of choosing.
A kind of method of detection candidate quantity for reducing object identifying method is proposed, wherein this method includes following step
It is rapid:
By using the score for accordingly distributing to detection candidate, the threshold value column from the list of evaluation detection candidate
Table；
Delete detection candidate, wherein by have be less than from threshold list the score of selected threshold value detection candidate from
It is deleted in list, particularly as long as it is not above the maximum distance with the detection candidate for calculating threshold value.
Detection candidate can be regarded as the image-region with finite size for wherein mapping searched for feature combination.Figure
It can be regarded as size less than the region by camera captured image as region is especially.Detection candidate particularly can be corner angle
Region candidate person, wherein mapping out the picture material of the feature with two seamed edges for being merged into corner angle.List may include image
Or image local area all detection candidates.The list may include various types of detection candidates.Evaluation is appreciated that
For the consistent degree with one group of test object template (such as with its gray value or the variable being derived there).Score can be indicated to knowledge
The confidence level of evaluation or the identification of other quality.By score or scoring, the detection candidate of list can be ranked up.Threshold value can
To be the percentage of score.It can be each score, particularly to be assigned each candidate threshold value of score, to obtain
Threshold list.Different types of detection candidate can be separated and be handled.
Can be removed from the list following detection candidate, i.e., these detection candidates be located at part and/or pass through definition
In the detection candidate of threshold value and/or the selection region for being determined or being determined by the position of detection candidate and/or score.The selection
Region can be defined into the maximum distance of detection candidate.In other words, selection region can limit the influence of single detection candidate
Region.
Selection region can be defined by the radius around detection candidate and/or with the preset distance of detection candidate.It can
Round selection region is quickly and easily set.It also can define the selection region of other forms by using other range measurements.
This method may include filtration step, wherein filtering out the detection candidate of overlapping from list.The detection of overlapping is candidate
Person is likely to be related to identical corner angle.The detection candidate with highest score can be used in filtering.
Different object types can be individually performed the step of determining and deleting.For example, can handle apart from each other not
With the detection candidate of corner angle type, such as upper left, upper right, bottom right and/or lower-left.By separated object type, can for compared with
Small object prevents from mistakenly deleting the corner region candidate for identifying poorly.
It can be firstly for the detection candidate threshold value in list with highest score.It in other words, can be from highest
The detection candidate of score starts.The step of this method, may be repeated.By from highest score to the carry out side of lower score
Method can save many operational capabilities, because also very high in higher score lower threshold value, thus by being disposably removed from the list perhaps
More detection candidates.
This method for example can for example be realized in the controller with the mixed form of software or hardware or software and hardware.
The method proposed be particularly suitable in information system for driver in motor vehicle or driver assistance system from
The scope of dynamic Traffic Sign Recognition.Thus image can be recorded by means of arranging the video camera towards direction of travel in the car
Signal or vision signal, and can check the traffic sign for including in recorded picture signal or vision signal.It is handed over existing
In the case where logical mark, these traffic signs can be shown, so as to for example to driver with information.In principle it is also possible to consider
It is driver assistance function to be controlled according to the traffic sign identified, such as speed is restricted to be indicated maximum allowable
Speed.
A kind of method as described above is proposed thus, wherein utilizing figure of the camera record from motor vehicle ambient enviroment
As signal or vision signal, the camera arrangements are in the motor vehicle or at the motor vehicle, wherein by using any of the above institute
Whether the method stated is come comprising traffic sign in check image signal or vision signal, and wherein when comprising traffic sign,
Output represents the signal of identified traffic sign.In addition, the scheme introduced herein also proposed a kind of device, it is configured to
The step of being executed in corresponding equipment, controlling or implement the variant schemes for the method introduced herein.Through the invention this
The implementation modification of device form can also quickly and efficiently realize the purpose that the present invention is based on.
For this purpose, the device can include at least one arithmetic element for handling signal or data；For store signal or
At least one storage unit of data；To at least one of sensor or actuator interface, for reading sensor from sensor
Data-signal or control signal are output to actuator by signal；And/or at least one communication interface, for reading or exporting
The data being embedded in communication protocol.Arithmetic element for example can be signal processor, microcontroller etc., and wherein storage unit can
To be flash memory, EEPROM or magnetic memory cell.Communication interface can be designed to wirelessly and/or wiredly read or export number
According to wherein the communication interface that can be read or export cable data for example can electrically or optically be read from corresponding data line
Data are output data in corresponding data line.
Here, device can be regarded as processing sensor signal and the accordingly electricity of output control signal and/or data-signal
Gas equipment.The device can have the interface that can be constructed based on hardware and/or software.In hardware based make, interface
Such as it can be a part comprising the various functions of device of so-called ASIC system.However also it is possible that interface can be
The integrated circuit of itself is at least partly made of discrete component.In software-based make, interface can be example
Such as with other software module and deposit software module on a microcontroller.
Further it is proposed that a kind of image processing system, has according to the scheme herein proposed for reduced device.
The device and image processing system especially can be in the scopes of automatic landmark identification or automatic traffic landmark identification
In in information system for driver or driver assistance system.Here, picture signal or vision signal to be assessed are by installing
The preferred video camera oriented towards direction of travel in the car or at vehicle records, and is transported to the signal processing of device
Portion.The signal processing part is suitable for identifying the traffic sign in recorded picture signal or vision signal by using the above method
And export the signal for the traffic sign for indicating identified.In order to identify traffic sign, such as it can refer to have and be stored therein
Traffic sign pattern image data base.Indicate the signal of identified traffic sign can be used for controlling in head up display or
The display of such as traffic sign symbol in vehicle on another display element.In addition, indicating the signal of identified traffic sign
It can also be used to control driver assistance function, such as the automatic rate limitation when identifying maximum permission speed.
It is also advantageous in that computer program product or computer program with program code, the program code are storable in
On machine readable carrier or storage medium (such as semiconductor memory, harddisk memory or optical memory) and for holding
The step of going, implementing and/or control the method according to one of above embodiment, especially when the program product or program are being counted
When being run on calculation machine or device.
Detailed description of the invention
The embodiment for the scheme introduced herein is shown in the accompanying drawings and is illustrated in detail in the following description.Its
In:
Fig. 1 shows the block diagram for reduced device according to one embodiment；
Fig. 2 shows the schematic diagrames according to the image with corner angle candidates and object of one embodiment；
Fig. 3 shows the schematic diagram of the image with detection candidate according to one embodiment；
Fig. 4 shows the schematic diagram of the image with detection candidate score according to one embodiment；And
Fig. 5 shows the flow chart of the method according to one embodiment.
Specific embodiment
In the explanation below to advantageous embodiment of the invention, for the similar member of effect shown in different figures
Part uses the same or similar appended drawing reference, wherein the repeated explanation to these elements is omitted.
Fig. 1 shows the block diagram for reduced device 100 according to one embodiment.Device 100 is the figure of vehicle 104
As the component part of processing system 102.Here, image processing system 102 further includes at least one video camera 106 and for identification
Device 108.Video camera 106 detects a part of the environment of vehicle 104 in its detection range and is imaged on image 110
In.Video camera 106 can also provide vision signal 110 from a series of images.Device 108 for identification identifies image by corner angle
There is the object 112 of corner angle shown in 110.
For this purpose, there is rib by using corner angle searching algorithm search image 110 in the search equipment 114 of device 108
The image-region 116 of corner characteristics.Wherein map corner angle feature image-region 116 be selected as detection candidate 118 and
Determine the position of corner angle.The score of the quality of corner angle detection or position detection is indicated for each detection candidate 118 distribution
120.It is assigned higher score 120 compared with the detection candidate 118 of the Supreme People's Procuratorate's mass metering here, having, and there is lower detection matter
The detection candidate 118 of amount is assigned lower score.
The list 122 of detection candidate 118 and its score 120 is transferred to the device 100 for reduction.In device 100
Really in locking equipment 124, to detect 118 allocation threshold 126 of candidate, which depends on the score of detection candidate 118
120.Here, score 120 is higher, then threshold value 126 is higher.In sweep equipment 128, deleting from list 122 has less than threshold
The detection candidate 118 of the score 120 of threshold value in 126 list of value, particularly as long as it is no more than and is considered for calculating
The maximum distance of the detection candidate of threshold value.List 122 becomes shorter as a result,.
When the position for detecting candidate 118 complies with standard, detection candidate 118 in the list 122 of shortening with
Subject area 132 is assigned to by using geometric standard in the distributing equipment 130 of the device 108 of identification.It is every in list
A detection candidate is all assigned threshold value, thus automatically generates above-mentioned threshold list.Delete unit be for example based not only on threshold value and
And detection candidate is deleted based on the distance between threshold list and detection candidate, therefore for example delete and meet the following conditions
Each detection candidate, i.e., its simultaneously be less than the threshold value in list and lower than with define respective threshold candidate at a distance from.
This can for example be effectively realized in the following manner, i.e., be ranked up in descending order to score list, for each threshold value and correspondence
Candidate delete score list, update the list and repeat whole process.
In one embodiment, selection region is distributed in around the position of detection candidate 118 in sweep equipment 128.
If other detection candidates 118 in the selection region have the score 120 of the threshold value 126 lower than detection candidate 118,
They are deleted from list 122.
In one embodiment, before threshold value 126, the detection candidate 118 of overlapping is merged into single inspection
Survey candidate 118.Here, obtained detection candidate 118 has the position of the highest detection candidate 118 of score 120.
Fig. 2 shows the schematic diagrames according to the image 110 with corner angle candidate 118 and object 112 of one embodiment.
The object 112 drawn is rectangle traffic sign 112 as shown in Figure 1.Here, traffic sign 112 is circular traffic sign 200
The additional guideboard 112 of lower section.There are four the corner angle being differently directed for the tool of rectangle traffic sign 112.For the traffic mark of horizontal orientation
Will 112, these corner angle are referred to alternatively as upper right, upper left, lower-left and bottom right.Here, being identified by different detection candidates 118
The corner angle being differently directed.It, must be near the corner angle of left side in the rectangular object 112 of perfect imaging due to geometric Framework condition
Find corresponding right side corner angle in right side.Equally, downside corner angle must be found in the lower section of upside corner angle.Therefore, in object search
When region 132, for example, two detection candidates 118 for bottom right corner angle can only belong to two different objects 112 or two
One of them of a detection candidate 118 is misidentified.
Reduce the quantity of error detection by the scheme introduced herein because more insecure detection candidate 118 due to
Its score is lower to be just deleted before object search region 132.
In the method for the angle the n geometric object 112 in image 110 for identification, whole in known image (part)
Wicket is slided on a image 110, wherein determining whether by means respectively of known method comprising object corner region.To n
Each of object corner angle repeat the process.Use color and geometry of searched for object corner region etc. special herein
Sign, to determine the window evaluation for finally obtaining assessment score.Learn these features by means of practicing data.This n detection process
Result be each object corner angle candidate corner angle position list.It is now based on these positions and corresponding score and is applicable in
Object candidates 132 are generated in the geometrical boundary condition of target object 112 to be identified.This method is proved to hand over for additional
The problem of logical landmark identification, is very reliable.Here, goal object 112 is rectangle.This method can be dedicated for main traffic
Indicate the image section of lower section.
In general, this method can be used to identify the traffic panel 112 in complete image 110.Know here, being similar to additional mark
There is not can be changed the rectangle of aspect ratio.Here, based on many knots being imaged in image 110 especially occurred in city
Structure, this method generate larger numbers of wrong report corner region candidate 118.
Fig. 3 shows the schematic diagram of the image 110 with detection candidate 118 according to one embodiment.For example, as schemed
Shown in 1, by the camera record image 110 of vehicle.Image 110 is acquired through the front windshield of vehicle and shows vehicle
A part of the vehicle-periphery in front.Here, the vehicle is in the running on expressway for having two orientation lanes.Due under
Rain, visibility are restricted.
The detection candidate 118 of different type corner angle is marked in the picture.The many detected in candidate 118 is mistake
And do not mark the corner angle in image 110.For example, many detection candidates are marked in the nimbus region shown
118.However, these detection candidates 118 have very low score, this is indicated by the thickness at the edge of each indicia framing.
Rectangle guideboard 112 can be seen in right side roadside.The detection candidate 118 with highest score in the lower right corner is marked
The lower right corner of guideboard 112.The scheme introduced herein by application, can eliminate the detection candidate 118 of many mistakes, because
Very indiscernible another lower right corner very possible and that another way board is not present around the lower right corner of reliable recognition.
Fig. 4 shows the partially schematic of the image 110 with the score 120 for detecting candidate 118 according to one embodiment
Figure.The vehicle camera record of image 110 for example as shown in Figure 1.As shown in figure 3, the front windshield through vehicle obtains figure
As 110.The Local map shows the rectangle guideboard 112 at portal frame 400 on a highway.
As shown in figure 3, showing the detection candidate 118 of corner region.Lower-left detection candidate is illustrated only herein
118.Other than the position of detection candidate 118, the score 120 of detection candidate 118 also is shown as numerical value.
At guideboard 112, here it is shown that three detection candidates 118 in the lower left corner.Here, a practical left side for guideboard 112
The detection candidate 118 of inferior horn is cited as 500 points of highest score 120 herein.The detection in the highway symbol lower left corner is waited
The person of choosing 118 is cited as 200 points of lower score.The detection candidate 118 in the lower left corner of the alphabetical D of place name " Dortmund " is commented
For 50 points of even lower score 120.
500 points of highest score 120 sets the height of threshold value herein.Here, by the score 120 multiplied by exemplary selection
0.25 factor, with obtain 125 threshold value.
In order to delete detection candidate 118, the detection candidate 118 around practical corner angle is distributed circular selection region
402.The detection candidate 118 of D is located in selection region 402 and has the score 120 lower than threshold value, therefore the quilt from list
It deletes.
The detection candidate 118 of highway symbol is similarly positioned in selection region 402, but has the score higher than threshold value
120, therefore will not be deleted from list.
Other unshowned detection candidates except selection region 402 are not considered when removed.
Present the variable inhibition to the wrong report object detection candidate 118 in partial image region 402.
The method (hereinafter referred to as rNMS (the non-maximum suppression based on radius)) introduced herein is for inhibiting wrong report candidate
118, for the result of the standard object recognition methods applied to image 110.This be a kind of its validity by experience by means of
Successfully test the algorithm being verified.In principle, it can score with the form output with score 120 so as to candidate to detection
All object identifying methods that person 118 is ranked up are used together.Due to its special characteristic, object is identified with entirety is used
Method is compared, and rNMS combination corner angle recognition methods provides obvious better result.
The rNMS method introduced herein reduces the quantity of wrong report 118.For this purpose, in each object corner region identified
In the neighborhood of O 118, the every other object that those assessment scores 120 are lower than the threshold value selected relative to the score 120 of O is deleted
Corner angle candidate 118.This is proved by following thought: the highest candidate 118 of scoring in image-region 402 has determined
Due to the accessible largest score 120 of characteristics of image in the image-region 402, so that every other candidate 118 only can reach
A part of the score 120.
The advantages of algorithm idea, is the opposite selection of its higher speed and locally applied score threshold.As a result,
The quantity of the corner region candidate 118 of wrong report can for example halve.This method can also be used in the object inspection of whole detection object 112
Survey device.Since the quantity wherein reported by mistake is usually much lower, can less exclude to report by mistake.
Typical problem in Object identifying is, eliminates all that " non-maximum " from a large amount of detection candidate frames 118
Value.This problem inhibited by non-maximum value, i.e. NMS is solved.Here, always referenced to object identifying method being each inspection
The existing assessment score that candidate frame 118 determines is surveyed to define non-maximality.
The object identifying method of whole identification target object can generate the candidate of many overlappings in one step.
In the global NMS based on score, the score of all candidates 118 and fixed pre-selected threshold can be compared
Compared with, and eliminate all candidates 118 that score is lower than the threshold value.In contrast, using opposite in the scheme introduced herein
There is the selected threshold value of score in maximum.
In the NMS based on overlapping, when two candidates 118 are with certain percentage overlapping, then can delete has
The candidate 118 of lower score.
This method can " greediness " be performed, i.e., by only comparing two candidates 118 every time and deleted immediately,
Or be globally performed, i.e., by comparing every two candidate 118 to all candidates and being deleted as final step
It removes.
It is average in radius to be parameterized will test the center of frame 118 that mean deviation algorithm can be used, thus should
All candidates 118 in radius merge into an output candidate.
It in the detection process iteratively may include about determining during object detection in alternative manner
Detection candidate 118 information, this may cause extremely complex method.
In contrast to this, whole image 110 is first checked for using at least one detector in method described herein,
Then image-region 402 is defined for each detection 118, wherein by the phase of already existing other detection candidates 118 and part
Threshold value is compared.Threshold value and the relativity and conventional method of the detection 118 of definition image-region 402 are significantly different.
The rNMS introduced herein can be used as the supplement to NMS or average shift NMS based on overlapping.Here, can also eliminate
Nonoverlapping candidate 118.
In the scheme introduced herein, the content of candidate image area 118 will not be rethought again.Thus this method can be fast
Speed carries out.
The new method introduced herein is formed in the intermediate steps of algorithm in the detection of object 112, and especially suitable for inspection
The angle n geometric object 112 in altimetric image 110.It is identifying the detection candidate 118 in image 110 and is passing through based on overlapping
NMS inhibit all overlapped candidates 118 after, be used in this introduction method.For each object 112
Candidate 118 is iterated by all detection candidates 118.Therefore, in the case where corner region detector 118, to every
A image 110 executes n times this method.The score P (i, j) 120 of i-th of candidate based on j-th of corner angle is calculated threshold value S (j)
=c (j) * P (i, j) and by all candidates in the radius r (j) 402 around candidate 118 of itself and j-th corner angle
118 score 120 is compared.If the score 120 of observed candidate 118 is less than S (j), candidate 118 is deleted
(greedy process variant).(global variant schemes) is performed alternatively, deleting and can be used as final step.If this method is used for
Whole object 112, then j is consistently equal to 1.
Radius r (j) 402 is parameter to be selected.Here, radius 402 may be, for example, the 1/5 of picture traverse.If by r
(j) it 402 is chosen so as to the radius 402 whole image 110 is all covered for any position, then the method introduced is equivalent to pair
The global of corner angle candidate 118 with relative threshold inhibits.
Parameter c (j) can be fixedly selected manually, or can be determined based on the result of practice data group.
This method is very quick, because only executing considerably less arithmetical operation, and can greatly reduce wrong report corner region
The quantity of candidate 118.It can be used as to the supplement of other methods for non-maximum suppression.
The NMS method introduced herein can also be used in arbitrary method for checking object, such as in pedestrian's identification and/or traffic mark
Will identifies field.However, corner region detection 118 is only intermediate steps in the method for the corner region of identification object 112,
It may generate more wrong reports 118 than the method for direct detected target object.
Fig. 5 shows the method 500 of the detection candidate quantity for reducing object identifying method according to one embodiment
Flow chart.For example, can be as shown in Figure 1 for executing method 500 on reduced device.This method 500 includes determining 502
The step of and delete 504 the step of.In the step of determining 502, by using accordingly distribute to detection candidate score from
Threshold value list in the list of evaluation detection candidate.In the step of deleting 504, score is removed from the list less than threshold value
The detection candidate of threshold value in list, particularly as long as its be no more than with the detection candidate for calculating threshold value it is maximum away from
From.
If one embodiment includes "and/or" conjunction between fisrt feature and second feature, this can be read as,
The embodiment not only with fisrt feature but also has second feature according to a kind of embodiment, and according to another embodiment
Only there is with fisrt feature or only second feature.
Claims (11)
1. a kind of for reducing the method (500) of the quantity of the detection candidate (118) of object identifying method, wherein the method
(500) the following steps are included:
By using the score (120) for accordingly distributing to detection candidate (118), from evaluation detection candidate (118)
(502) threshold value (126) list is determined in list (122)；
(504) detection candidate (118) is deleted, wherein deleting score (120) from the list (122) is less than the threshold value
(126) the detection candidate (118) of the threshold value in list, particularly as long as it is no more than and the detection for calculating the threshold value
The maximum distance of candidate (118).
2. according to the method for claim 1 (500), wherein in the deletion (504) the step of, from the list
(122) following detection candidate (118) is deleted in: detection candidate (118) is showing detection candidate (118)
It is located at the selection region (402) determined by the position and/or score (120) of detection candidate (118) on image (110)
It is interior.
3. according to the method for claim 2 (500), wherein in the deletion (504) the step of, the selection region
(402) determine by the radius around detection candidate (118) and/or with the preset distance for detecting candidate (118)
Justice.
4. method according to any of the preceding claims (500), includes the steps that filtering, wherein from the list
(122) the detection candidate (118) of overlapping is filtered out in.
5. method according to any of the preceding claims (500), wherein the step of determination (502) and described deleting
The step of except (504), is respectively executed different object types.
6. method according to any of the preceding claims (500), wherein in the determination (502) the step of, it is first
It is first the detection candidate (118) in the list (122) with maximum score (120) from the threshold value (126) list
Threshold value.
7. method according to any of the preceding claims, wherein coming from motor vehicle ambient enviroment using camera record
Picture signal or vision signal, the camera arrangements are in the motor vehicle or at the motor vehicle, wherein by using
Whether method according to any of the preceding claims checks in described image signal or vision signal comprising traffic
Mark, and wherein when comprising traffic sign, output indicates the signal of identified traffic sign.
8. a kind of device (100), suitable for executing method according to any of the preceding claims in corresponding unit
(500) the step of.
9. a kind of image processing system (102) has according to claim 8 for reduced device (100).
10. a kind of computer program is adapted for carrying out method according to any of the preceding claims (500).
11. a kind of machine readable storage medium, is stored with calculating according to claim 10 on said storage
Machine program.
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE102017212426.1 | 2017-07-20 | ||
| DE102017212426.1A DE102017212426A1 (en) | 2017-07-20 | 2017-07-20 | Method and apparatus for reducing a number of detection candidates of an object recognition method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109284665A true CN109284665A (en) | 2019-01-29 |
| CN109284665B CN109284665B (en) | 2022-02-08 |
Family
ID=64951944
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810789967.0A Active CN109284665B (en) | 2017-07-20 | 2018-07-18 | Method and apparatus for reducing number of detection candidates for object recognition method |
Country Status (3)
| Country | Link |
|---|---|
| CN (1) | CN109284665B (en) |
| DE (1) | DE102017212426A1 (en) |
| FR (1) | FR3069362B1 (en) |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100303342A1 (en) * | 2009-06-02 | 2010-12-02 | Yahoo! Inc. | Finding iconic images |
| US20120219222A1 (en) * | 2011-02-24 | 2012-08-30 | Ahmet Mufit Ferman | Methods and Systems for Determining a Document Region-of-Interest in an Image |
| CN103034844A (en) * | 2012-12-10 | 2013-04-10 | 广东图图搜网络科技有限公司 | Image identification method and device |
| US20130216118A1 (en) * | 2010-11-05 | 2013-08-22 | Peter Keith Rogan | Centromere Detector and Method for Determining Radiation Exposure From Chromosome Abnormalities |
| CN103310469A (en) * | 2013-06-28 | 2013-09-18 | 中国科学院自动化研究所 | Vehicle detection method based on hybrid image template |
| CN106815604A (en) * | 2017-01-16 | 2017-06-09 | 大连理工大学 | Method for viewing points detecting based on fusion of multi-layer information |
| CN106845458A (en) * | 2017-03-05 | 2017-06-13 | 北京工业大学 | A kind of rapid transit label detection method of the learning machine that transfinited based on core |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP2709038A1 (en) * | 2012-09-17 | 2014-03-19 | Thomson Licensing | Device and method for detecting the presence of a logo in a picture |
-
2017
- 2017-07-20 DE DE102017212426.1A patent/DE102017212426A1/en active Pending
-
2018
- 2018-07-17 FR FR1856566A patent/FR3069362B1/en active Active
- 2018-07-18 CN CN201810789967.0A patent/CN109284665B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20100303342A1 (en) * | 2009-06-02 | 2010-12-02 | Yahoo! Inc. | Finding iconic images |
| US20130216118A1 (en) * | 2010-11-05 | 2013-08-22 | Peter Keith Rogan | Centromere Detector and Method for Determining Radiation Exposure From Chromosome Abnormalities |
| US20120219222A1 (en) * | 2011-02-24 | 2012-08-30 | Ahmet Mufit Ferman | Methods and Systems for Determining a Document Region-of-Interest in an Image |
| CN103034844A (en) * | 2012-12-10 | 2013-04-10 | 广东图图搜网络科技有限公司 | Image identification method and device |
| CN103310469A (en) * | 2013-06-28 | 2013-09-18 | 中国科学院自动化研究所 | Vehicle detection method based on hybrid image template |
| CN106815604A (en) * | 2017-01-16 | 2017-06-09 | 大连理工大学 | Method for viewing points detecting based on fusion of multi-layer information |
| CN106845458A (en) * | 2017-03-05 | 2017-06-13 | 北京工业大学 | A kind of rapid transit label detection method of the learning machine that transfinited based on core |
Non-Patent Citations (1)
| Title |
|---|
| NIRMALA RAMAKRISHNAN ET AL: "《Automated Thresholding for Low-Complexity Corner Detection》", 《2014 NASA/ESA CONFERENCE ON ADAPTIVE HARDWARE AND SYSTEMS》 * |
Also Published As
| Publication number | Publication date |
|---|---|
| DE102017212426A1 (en) | 2019-01-24 |
| FR3069362B1 (en) | 2023-01-06 |
| CN109284665B (en) | 2022-02-08 |
| FR3069362A1 (en) | 2019-01-25 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US10970566B2 (en) | Lane line detection method and apparatus | |
| US9047518B2 (en) | Method for the detection and tracking of lane markings | |
| US9467645B2 (en) | System and method for recognizing parking space line markings for vehicle | |
| US8050459B2 (en) | System and method for detecting pedestrians | |
| JP4380838B2 (en) | Video image automatic road sign recognition method, road sign automatic recognition device, and road sign automatic recognition program | |
| US9076047B2 (en) | System and method for recognizing parking space line markings for vehicle | |
| CN104376297A (en) | Detection method and device for linear indication signs on road | |
| CN112598922B (en) | Parking space detection method, device, equipment and storage medium | |
| WO2008130219A1 (en) | Method of and apparatus for producing road information | |
| US20100202661A1 (en) | Moving object detection apparatus and computer readable storage medium storing moving object detection program | |
| CN106951898B (en) | Vehicle candidate area recommendation method and system and electronic equipment | |
| CN107452035B (en) | Method and apparatus for analyzing lane line image and computer readable medium thereof | |
| US20160055382A1 (en) | Vehicle discrimination apparatus | |
| JP6818626B2 (en) | Vehicle type discrimination device, vehicle type discrimination method, and vehicle type discrimination system | |
| JP2013057992A (en) | Inter-vehicle distance calculation device and vehicle control system using the same | |
| CN112307840A (en) | Indicator light detection method, device, equipment and computer readable storage medium | |
| CN106326821B (en) | The method and device of License Plate | |
| EP3809399A1 (en) | Map generation device and map generation system | |
| CN108319931A (en) | A kind of image processing method, device and terminal | |
| CN109102026A (en) | A kind of vehicle image detection method, apparatus and system | |
| JP2006201817A (en) | Method and device for discriminating vehicle body color | |
| CN109344829A (en) | A kind of Train number recognition method and device of High Speed Railway Trains | |
| CN109284665A (en) | Method and apparatus for reducing the detection candidate quantity of object identifying method | |
| JP2014126887A (en) | License plate determination device | |
| CN107844749B (en) | Road surface detection method and device, electronic device and storage medium |
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 |