CN109409360A - A kind of multiple dimensioned image object detection method and detection system - Google Patents
A kind of multiple dimensioned image object detection method and detection system Download PDFInfo
- Publication number
- CN109409360A CN109409360A CN201811125321.9A CN201811125321A CN109409360A CN 109409360 A CN109409360 A CN 109409360A CN 201811125321 A CN201811125321 A CN 201811125321A CN 109409360 A CN109409360 A CN 109409360A
- Authority
- CN
- China
- Prior art keywords
- harr
- scale
- harr feature
- integrogram
- image
- 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.)
- Pending
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/255—Detecting or recognising potential candidate objects based on visual cues, e.g. shapes
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Image Analysis (AREA)
Abstract
A kind of multiple dimensioned image object detection method of the invention, including S1: obtaining the harr feature original pattern of target to be detected；S2: the pixel value of complete image to be detected is obtained；S3: the integrogram of complete image to be detected is obtained；S4: traversing each region of image, and upper left endpoint deviation post in the picture of each region based on Harr feature original pattern, one's respective area judges whether to meet Harr feature, record testing result；S5: judging all positions for meeting harr feature, is confirmed whether to detect the position of object to be measured and object to be measured.Compared with prior art, the application calculates in calculating process without carrying out the scaling of image, avoids the influence being distorted to detection performance.And the integrogram of storage complete image is only needed, reduce memory space.Simultaneously because this process is only calculated once integrogram, computational complexity is greatly simplified.
Description
Technical field
The present invention relates to technical field of image processing, specifically a kind of multiple dimensioned image object detection method and inspection
Examining system.
Background technique
Present image target detection has been widely used in Face datection, intelligent monitoring, the fields such as robot navigation.It is described
Image object detection refers to the technology for judging that the figure of target to be detected whether there is in some full picture or image.It is logical
Cross it is further improve and application, image object detection not only may determine that target to be detected whether there is, can also obtain to
Detect specific location of the target in designated pictures.The further application of image detection may also act to the images such as image registration
Process field.The application more extensively sent out wherein is obtained with the technology that the image object of Harr feature detects.Harr feature is by one
A Window (can be considered as the two-dimensional pixel matrix that a size defines), is divided into more than two parts, to this inspection
The pixel value of each section surveyed in window is summed respectively, whether is greater than thresholding by the ratio of its each section, come be made whether containing
The judgement of target.
It should be noted that the size of usually this Window is just pre-defined before inputting complete image
's.Those skilled in the art have found that target object size to be detected can exist and feature predetermined in practical applications
The unmatched situation of window, so that detection failure.
Further, those skilled in the art usually using multi-scale transform method, to solve " target pair to be detected
As size can exist and the unmatched problem of characteristic window predetermined ".
It can be summarized in existing multiple dimensioned image object detection method are as follows:
A.step0: the Harr feature original pattern of target to be detected is obtained.
A.Step1: the image data (pixel value) of complete image to be detected is obtained.
A.Step2: the change of scale information pre-defined according to some, to technologies hands such as complete image application interpolation
Duan Jinhang scaling.
A.Step3: integrogram calculating is carried out to the complete image after change of scale.
A.Step4: traversing each detection window of image, and each detection window is based on Harr feature original pattern, one's respective area
The deviation post of upper left endpoint in the picture calculates Harr feature, judges whether to meet Harr feature, records testing result.
Whether A.Step5: completing all predefined change of scale, if it is not complete, return A.Step2, otherwise after
Continue to A.Step6.
A.Step6: based on strategy, all positions for meeting Harr feature are judged.It is confirmed whether to detect to be measured
The position of target and object to be measured.
Although existing solution, solve " target object size to be detected can exist and feature predetermined
The unmatched problem of window ".
But due to needing to carry out complete image scaling, (this process industry have been recognized that for) can make that there are image mistakes
Very, so affecting the information of original complete image, Harr feature can be detected performance is brought to influence in this way.
On the other hand, due to be needed same to the continuous scaling of original complete image, so in entire treatment process
When storage complete image and transformed image, this has certain limitation to memory space.
In addition, scaling will recalculate integrogram every time, calculated so needing to calculate multiple interpolation and integrogram, fortune
It is also very big to calculate consumption.
Summary of the invention
The technical problem to be solved by the present invention is to how reduce memory consumption and operation consumption in image processing process.
To achieve the goals above, technical scheme is as follows:
A kind of multiple dimensioned image object detection method, including
S1: the harr feature original pattern of target to be detected is obtained；
S2: the pixel value of complete image to be detected is obtained；
S3: the integrogram of complete image to be detected is obtained；
S4: each region of integrogram, upper left endpoint of each region based on Harr feature original pattern, one's respective area are traversed
Deviation post in the picture judges whether to meet Harr feature, records testing result；
S5: judging all positions for meeting harr feature, is confirmed whether to detect object to be measured and mesh to be measured
Target position.
Preferably, the step S1 specifically: obtain the Harr feature of target to be detected, harr feature is that window is a height of
H0, width W0, Harr feature are S；
Wherein, S Harr feature of Harr (s) expression, the pixel value of dat (i, j) expression detection block internal coordinate (i, j), i <
W0, j < H0, s=0,1,2 ... S；
As Harr (s) > th (s), then it is assumed that the region meets Harr feature, and wherein th (s) is threshold value.
Preferably, the step S2 specifically: camera collects high H, and the picture of wide W is translated into gray value,
The gray value of each position is denoted as:
Imag (n, m), wherein n=0,1,2..., H-1, m=0,1,2..., W-1.
Preferably, the step S3 specifically: integral diagram data I (n, m) is obtained, wherein
N=0,1,2...H-1, m=0,1,2...W-1
Preferably, scanning search mode in the step S5 are as follows: integrogram is rectangle, detection window is rectangular window, from integral
Scheme any angle to start, the corresponding angular vertex of hough transform window is slided in entire integrogram, all in integrogram until having slided
Position；Wherein the size of each window to be detected is defined as original Harr Window size multiplied by maximum amplification: scale
(k), k=1,2 ..., K, K are natural number；So each window size to be detected is W0 × max (scale) × H0 × max
(scale)；
Each detection window is to the calculation of the Harr feature of different scale, with the position currently slided into
(offset1, offset2) is detection window vertex position, with the position and the sub- boundary point of graph of original Harr feature and
Scale determines the integrogram numerical value used, and the calculating of Harr feature, warp are calculated according to intrinsic subgraph and integral transform mode
It is as follows to cross the transformed harr feature calculation of scale:
Wherein (offset1, offset2) is position coordinates of the respective vertices of detection window in integrogram, value range
Respectively offset1=0,1,2 ..., W-1, offset2=0,1,2 ..., H-1.
Each Harrscale (s, k) of each detection zone is compared with respective threshold scale (k) × th1, if it is greater than door
Limit, then record respective vertices position (offset1, offset2) and width and elevation information, i.e. and W0 × scale (k), H0 ×
scale(k)。
The present invention also provides a kind of multiple dimensioned image object detection systems, applied to above-mentioned detection method, comprising:
Harr feature training module obtains harr feature to training；
Image collection module, to obtain the pixel value of complete image to be detected；
Integrogram conversion module, complete image is converted into integrogram；
Image scaling module, to carry out scaling to complete image according to the change of scale information pre-defined；
Spider module, traversing each region of image, each region is based on Harr feature original pattern, one's respective area
The deviation post of upper left endpoint in the picture judges whether to meet Harr feature, records testing result；；
Judgment module, to judge all positions for meeting harr feature；It is confirmed whether to detect object to be measured,
And the position of object to be measured.
Compared with the prior art, the present invention has the following beneficial effects:
The application calculates in calculating process without carrying out the scaling of image, avoids the influence being distorted to detection performance.
And in an implementation, it is only necessary to which the integrogram (size of integrogram and the size of original image are identical) for storing complete image reduces
Memory space.Simultaneously because this process is only calculated once integrogram, computational complexity is greatly simplified.
Specific embodiment
The effect of to make to structure feature of the invention and being reached, has a better understanding and awareness, to preferable
Embodiment is described in detail, as follows:
A kind of multiple dimensioned image object detection method, content are as follows:
Step 1. obtains the Harr feature original pattern of target to be detected.Harr feature is obtained by training, Harr feature
Common expression formula are as follows:
A kind of common situation of Set are as follows:
Set { s, n }=(i, j) | x [s, n] <=i <=x [s, n]+Wsubpart [s, n] -1, y [s, n] <=j <
=y [s, n]+Hsubpart [s, n] -1 },
Wherein, all i are both less than width W0, all j and are both less than high H0.This width is W0, the region of a height of H0 is known as detecting
Window.
Set { s, n } may be considered n-th of subgraph of s-th of feature.Weight [s, n] is the of s-th feature
N weight.X [s, n], x [s, n]+Wsubpart [s, n], y [s, n], y [s, n]+Hsubpart [s, n] are known as n subgraph
Boundary point.
Wherein, i, j are the coordinate in two-dimensional surface kind, and x [s, n], y [s, n] are considered the upper left in this sub-pattern region
Endpoint, Wsubpart [s, n] and Hsubpart [s, n] are considered the width and height of this sub-pattern.
It should be noted that in advance the obtained Harr feature of training be possible have it is multiple, it can be expressed as Harr (s),
Middle s=0,1,2 ..., S.
If there is multiple (S) Harr features, then the definition for meeting Harr feature can be both greater than door for multiple features
Limit th (s) (total S) then thinks that the region meets Harr feature.
Can also be defined as Harr feature meets feature with value.
Step 2: obtaining the pixel value of complete image to be detected；Camera collects high H, the picture of wide W, the present embodiment
It is middle to use high by 640, wide 480 picture, it is translated into gray value, the gray value of each position is denoted as:
Imag (n, m), wherein n=0,1,2...H-1, m=0,1,2...W-1.
Step 3. obtains the integrogram of complete image to be detected；According to conventional technical means, integral diagram data I is obtained
(n, m), wherein
N=0,1,2...H-1, m=0,1,2...W-1
Step 4. traverses each region of image, and based on position and region and zoom scale predetermined, (this is determined
Determine the coordinate information of Harr feature to be calculated) numerical value that takes out corresponding integrogram calculated.Each region is based on Harr
Feature original pattern, one's respective area upper left endpoint deviation post in the picture, judge whether to meet Harr feature, record inspection
Survey result；
Each region is according to Harr feature, scaling scale situation, location information of the one's respective area in complete image, corresponding part
Information carry out Harr feature judgement:
Scanning search mode is that since the picture upper left corner, hough transform window top left corner apex is whole in the present embodiment
It is slided in a image, until having slided all positions in image.
Wherein the size of each window to be detected is defined as original Harr detection window size multiplied by maximum amplification.
The multiple scale (k) for needing to change in the present embodiment, k=1 ... K, K are natural number, K=4 in the present embodiment,
Scale (1)=1, scale (2)=2, scale (3)=3, scale (4)=4.
So according to the concrete condition of the present embodiment, each window size to be detected are as follows:
W0*max(scale)×H0*max(scale)。
Each detection window is to the calculation of the Harr feature of different scale, with the position currently slided into
(offset1, offset2), value range are respectively 0...W-1,0...H-1, to detect window vertex position, with the position and
The sub- boundary point of graph of original Harr feature and scale determine the integrogram numerical value used, turn according to intrinsic subgraph and section point
Change mode calculates the calculating of Harr feature, by the transformed harr feature of scale.
I.e. transformed Harr feature are as follows:
If Harr feature have it is multiple, HarrScale (k) can with HarrScale (s, k) indicate, be s-th spy
K-th of transformation of sign,
SetScale { s, n, k } is k-th of transformation set of n-th of subgraph of s-th of feature, corresponding set mapping mode
Current sliding position is added multiplied by scale (k) in right boundary for the Harr feature of n-th of subgraph to s feature
(offset1,offset2)
SetScale { s, n, k }=
(i, j) | offset1+x [s, n] × scale (k)≤i
≤ offset1+ (x [s, n]+Wsubpart [s, n]) × scale (k) -1,
Offset2+y [s, n] × scale (k)≤j
≤ offset2+ (y [s, n]+Hsubpart [s, n]) × scale (k) -1 }
Offset1+x [s, n] can be denoted as x0；
Offset2+y [s, n] is denoted as y0；
Offset1+x [s, n]+Wsubpart [s, n] × scale (k) is denoted as x1；
Offset2+y [n]+Hsubpart [s, n] × scale (k) is denoted as y1
The calculating of associated quad figure is expressed as:
In based on the original image repeatedly existing method of scale transformation, the calculating of Harr feature can consider and only take
Scale=1, wherein offset1, offset2 are the position of the top left corner apex of detection window in the picture.
Each k of each detection zone, if Harrscale (s, k) compared with respective threshold scale (k) × th1 (s), if
Both greater than thresholding (be greater than s thresholding), then record endpoint location and Width x Height information (offset1, offset2, W0 ×
Scale (k), H0 × scale (k)).
Step 5. is based on strategy, judges all positions for meeting HarrScale feature.Be confirmed whether to detect to
Survey the position of target and object to be measured.
The present embodiment kind, use: judging the area of the overlapping region of multiple " positions (region) for meeting harr feature " is
It is no to meet the thresholding pre-defined, if it is satisfied, then thinking to detect target to be detected, and think the position with detection target
Setting (region) is the overlapping region.
This overlapping region may be considered the intersection in two regions by " position (region) for meeting harr feature ", or
The intersection in three regions of person or N number of region.This N value is also to be determined by the fact is predefined.
Embodiment 2
For the present embodiment the difference from embodiment 1 is that Harr feature is more specific, Harr feature is not trained but direct
Is defined as:
Harr feature is a kind of special case of general Harr feature in this implementation of step 1., only has a Harr feature, this
Harr feature is a height of H0 of window in embodiment, and width is left half of and right one side of something difference in the detection block of W0, in the present embodiment
H0=4, W0=4, i.e. detection window size are 4 × 4, calculation are as follows:
Dat (i, j) indicates the pixel value of detection block internal coordinate (i, j), which is relative to the opposite of detection block vertex
Coordinate, the pixel value are defined as gray value.There is also thresholding th1 for the feature of Harr1, the judgment criterion after calculating Harr1 feature
For, when the Harr1 > th1 is, then judgement meets Harr feature, is otherwise judged as and is unsatisfactory for Harr feature, and th1 value is-
25600~25600, value is 100 in the present embodiment.
Step 2. obtains the pixel value of complete image to be detected.Camera collects high H, the picture of wide W, the present embodiment
It is middle to use high by 640, wide 480 picture, it is translated into gray value, the gray value of each position is denoted as:
Imag (n, m), wherein n=0...H-1, m=0...W-1.
Step 3. obtains the integrogram of complete image to be detected；According to conventional technical means, integral diagram data I is obtained
(n, m), wherein
N=0,1,2...H-1, m=0,1,2...W-1
Whether each region Harr feature that step 4. successively calculates image meets the requirements:
Each region is according to Harr feature, scaling scale situation, location information of the one's respective area in complete image, corresponding part
Information carry out Harr feature judgement:
Scanning search mode is that since the picture upper left corner, hough transform window top left corner apex is whole in the present embodiment
It is slided in a image, until having slided all positions in image.
Wherein the size of each window to be detected is defined as original Harr detection window size multiplied by maximum amplification.
The multiple scale (k) for needing to change in the present embodiment, k=1 ... K, K are natural number, K=4 in the present embodiment,
Scale (1)=1, scale (2)=2, scale (3)=3, scale (4)=4.
So each window size to be detected is W0*max (scale) × H0*max according to the concrete condition of the present embodiment
(scale), as 16 × 16.
Each detection window is to the calculation of the Harr feature of different scale, with the position currently slided into
(offset1, offset2), value range are respectively offset1=0,1,2 ..., W-1, offset2=0,1,2 ..., H-1；
To detect window vertex position, the integrogram used is determined with the position and the sub- boundary point of graph of original Harr feature and scale
Numerical value, it is special by the transformed harr of scale according to the calculating that intrinsic subgraph and section divide transform mode to calculate Harr feature
Sign.
I.e. transformed Harr feature are as follows:
Divide transform mode according to intrinsic subgraph and section
HarrScale(k)
=I (offset1+4*scale (k) -1, offset2+2*scale (k) -1)
+ I (offset1, offset2)-I (offset1+4*scale (k) -1, offset2)
- I (offset1, offset2+2*scale (k) -1)
+ I (offset1+4*scale (k) -1, offset2+4*scale (k) -1)
+ I (offset1, offset2+2*scale (k))
- I (offset1+4*scale (k) -1, offset2+2*scale (k))
- I (offset1, offset2+4*scale (k) -1)
Wherein offset1, offset2 are the position of the top left corner apex of detection window in the picture.
Each HarrScale (k) of each detection zone is compared with respective threshold scale (k) × th1, if it is greater than thresholding,
Corresponding endpoint position (offset1, offset2) is then recorded, (W0 × scale (k), HO × scale (k)).
Step 5. is based on strategy, judges all positions for meeting HarrScale feature.Be confirmed whether to detect to
Survey the position of target and object to be measured.
The basic principles, main features and advantages of the present invention have been shown and described above.The technology of the industry
Personnel are it should be appreciated that the present invention is not limited to the above embodiments, the only present invention described in above-described embodiment and invention book
Principle, various changes and improvements may be made to the invention without departing from the spirit and scope of the present invention, these variation and
Improvement is both fallen in the range of claimed invention.The present invention claims protection scope by appended claims and
Its equivalent defines.
Claims (6)
1. a kind of multiple dimensioned image object detection method, it is characterised in that: including
S1: the harr feature original pattern of target to be detected is obtained；
S2: the pixel value of complete image to be detected is obtained；
S3: the integrogram of complete image to be detected is obtained；
S4: traversing each region of integrogram, and each region is being schemed based on the upper left endpoint of Harr feature original pattern, one's respective area
Deviation post as in judges whether to meet Harr feature, records testing result；
S5: judging all positions for meeting harr feature, is confirmed whether to detect object to be measured and object to be measured
Position.
2. the multiple dimensioned image object detection method of one kind according to claim 1, it is characterised in that: the step S1 tool
Body are as follows: obtain the Harr feature of target to be detected, harr feature is a height of H0 of window, and width W0, Harr feature is S；
Wherein, S Harr feature of Harr (s) expression, the pixel value of dat (i, j) expression detection block internal coordinate (i, j), i < W0, j <
H0, s=0,1,2 ... S；
As Harr (s) > th (s), then it is assumed that the region meets Harr feature, and wherein th (s) is threshold value.
3. the multiple dimensioned image object detection method of one kind according to claim 1, it is characterised in that: the step S2 tool
Body are as follows: camera collects high H, and the picture of wide W is translated into gray value, and the gray value of each position is denoted as:
Imag (n, m), wherein n=0,1,2..., H-1, m=0,1,2..., W-1.
4. the multiple dimensioned image object detection method of one kind according to claim 1, it is characterised in that: the step S3 tool
Body are as follows: integral diagram data I (n, m) is obtained, wherein
N=0,1,2...H-1, m=0,1,2...W-1.
5. the multiple dimensioned image object detection method of one kind according to claim 1, it is characterised in that: in the step S5
Scanning search mode are as follows: integrogram is rectangle, detection window is rectangular window, and since any angle of integrogram, hough transform window is corresponding
Angular vertex is slided in entire integrogram, until having slided all positions in integrogram；The wherein size of each window to be detected
Original Harr Window size is defined as multiplied by maximum amplification: scale (k), k=1,2 ..., K, K are natural number；So
Each window size to be detected is W0 × max (scale) × H0 × max (scale)；
Each detection window is to the calculation of the Harr feature of different scale, with currently slide into position (offset1,
Offset2 it is) detection window vertex position, is determined and used with the position and the sub- boundary point of graph of original Harr feature and scale
Integrogram numerical value, the calculating of Harr feature is calculated according to intrinsic subgraph and integral transform mode, it is transformed by scale
Harr feature calculation is as follows:
Wherein (offset1, offset2) is position coordinates of the respective vertices of detection window in integrogram, value range difference
For offset1=0,1,2 ..., W-1, offset2=0,1,2 ..., H-1.
Each Harrscale (s, k) of each detection zone is compared with respective threshold scale (k) × th1, if it is greater than thresholding, then
Record respective vertices position (offset1, offset2) and width and elevation information, i.e. W0 × scale (k), H0 × scale
(k)。
6. a kind of multiple dimensioned image object detection system, it is characterised in that: it is any described to be applied to the claims 1 to 5
Detection method, comprising:
Harr feature training module obtains harr feature to training；
Image collection module, to obtain the pixel value of complete image to be detected；
Integrogram conversion module, complete image is converted into integrogram；
Image scaling module, to carry out scaling to complete image according to the change of scale information pre-defined；
Spider module, to traverse each region of image, upper left of each region based on Harr feature original pattern, one's respective area
The deviation post of endpoint in the picture judges whether to meet Harr feature, records testing result；；
Judgment module, to judge all positions for meeting harr feature；It is confirmed whether to detect object to be measured, and
The position of object to be measured.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811125321.9A CN109409360A (en) | 2018-09-26 | 2018-09-26 | A kind of multiple dimensioned image object detection method and detection system |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811125321.9A CN109409360A (en) | 2018-09-26 | 2018-09-26 | A kind of multiple dimensioned image object detection method and detection system |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109409360A true CN109409360A (en) | 2019-03-01 |
Family
ID=65466151
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811125321.9A Pending CN109409360A (en) | 2018-09-26 | 2018-09-26 | A kind of multiple dimensioned image object detection method and detection system |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109409360A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919228A (en) * | 2019-03-08 | 2019-06-21 | 旺微科技(浙江)有限公司 | A kind of rapid detection method and device of target |
| CN109948626A (en) * | 2019-03-08 | 2019-06-28 | 旺微科技(浙江)有限公司 | A kind of object detection method and device |
| CN112016567B (en) * | 2020-10-27 | 2021-02-12 | 城云科技（中国）有限公司 | Multi-scale image target detection method and device |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559501A (en) * | 2013-10-25 | 2014-02-05 | 公安部第三研究所 | Vehicle sun visor detecting method and device based on image analysis |
| CN105046278A (en) * | 2015-07-17 | 2015-11-11 | 江苏物联网研究发展中心 | Optimization method of Adaboost detection algorithm on basis of Haar features |
| CN107145842A (en) * | 2017-04-19 | 2017-09-08 | 西安电子科技大学 | With reference to LBP characteristic patterns and the face identification method of convolutional neural networks |
-
2018
- 2018-09-26 CN CN201811125321.9A patent/CN109409360A/en active Pending
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103559501A (en) * | 2013-10-25 | 2014-02-05 | 公安部第三研究所 | Vehicle sun visor detecting method and device based on image analysis |
| CN105046278A (en) * | 2015-07-17 | 2015-11-11 | 江苏物联网研究发展中心 | Optimization method of Adaboost detection algorithm on basis of Haar features |
| CN107145842A (en) * | 2017-04-19 | 2017-09-08 | 西安电子科技大学 | With reference to LBP characteristic patterns and the face identification method of convolutional neural networks |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109919228A (en) * | 2019-03-08 | 2019-06-21 | 旺微科技(浙江)有限公司 | A kind of rapid detection method and device of target |
| CN109948626A (en) * | 2019-03-08 | 2019-06-28 | 旺微科技(浙江)有限公司 | A kind of object detection method and device |
| CN109919228B (en) * | 2019-03-08 | 2023-04-11 | 旺微科技(浙江)有限公司 | Target rapid detection method and device |
| CN112016567B (en) * | 2020-10-27 | 2021-02-12 | 城云科技（中国）有限公司 | Multi-scale image target detection method and device |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109544677A (en) | Indoor scene main structure method for reconstructing and system based on depth image key frame | |
| CN109409360A (en) | A kind of multiple dimensioned image object detection method and detection system | |
| CN107014294A (en) | A kind of contact net geometric parameter detection method and system based on infrared image | |
| CN106909885A (en) | A kind of method for tracking target and device based on target candidate | |
| CN106997459A (en) | A kind of demographic method split based on neutral net and image congruencing and system | |
| CN106920247A (en) | A kind of method for tracking target and device based on comparison network | |
| Huang et al. | Robust lane marking detection under different road conditions | |
| CN104992431B (en) | The method and device of multi-spectral image registration | |
| Chen et al. | Multiple object tracking using edge multi-channel gradient model with ORB feature | |
| Shalnov et al. | Convolutional neural network for camera pose estimation from object detections | |
| Wang et al. | Improving RGB-D SLAM accuracy in dynamic environments based on semantic and geometric constraints | |
| CN106651923A (en) | Method and system for video image target detection and segmentation | |
| CN109684909B (en) | Real-time positioning method, system and storage medium for target essential points of unmanned aerial vehicle | |
| Küçükyildiz et al. | Development and optimization of a DSP-based real-time lane detection algorithm on a mobile platform | |
| CN117611525A (en) | Visual detection method and system for abrasion of pantograph slide plate | |
| CN109961637A (en) | Vehicle detection apparatus and system based on more subgraphs fusion and significance analysis | |
| CN116863170A (en) | Image matching method, device and storage medium | |
| Tian et al. | An improved lane boundaries detection based on dynamic ROI | |
| Liu et al. | A visual SLAM method assisted by IMU and deep learning in indoor dynamic blurred scenes | |
| Zhang et al. | Real-time Lane Detection Method Based On Region Of Interest | |
| CN104182993B (en) | Target tracking method | |
| CN108428250A (en) | A kind of X angular-point detection methods applied to vision positioning and calibration | |
| CN113505806A (en) | Robot grabbing detection method | |
| Fang et al. | SLAM algorithm based on bounding box and deep continuity in dynamic scene | |
| Sang et al. | An improved checkerboard detection algorithm based on adaptive filters |
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 | ||
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190301 |