CN109741333A - A kind of improved object detection method, system and device - Google Patents
A kind of improved object detection method, system and device Download PDFInfo
- Publication number
- CN109741333A CN109741333A CN201811652762.4A CN201811652762A CN109741333A CN 109741333 A CN109741333 A CN 109741333A CN 201811652762 A CN201811652762 A CN 201811652762A CN 109741333 A CN109741333 A CN 109741333A
- Authority
- CN
- China
- Prior art keywords
- picture
- local
- detection
- region
- original
- 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
- 238000001514 detection method Methods 0.000 title claims abstract description 123
- 238000012360 testing method Methods 0.000 claims abstract description 24
- 238000000034 method Methods 0.000 claims abstract description 19
- 238000005516 engineering process Methods 0.000 claims abstract description 16
- 238000005549 size reduction Methods 0.000 claims abstract description 10
- 238000000605 extraction Methods 0.000 claims description 9
- 238000013527 convolutional neural network Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 6
- 230000004927 fusion Effects 0.000 claims description 5
- 230000003321 amplification Effects 0.000 claims description 3
- 230000005764 inhibitory process Effects 0.000 claims description 3
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 3
- 238000007689 inspection Methods 0.000 abstract description 7
- 238000013135 deep learning Methods 0.000 description 6
- 238000010586 diagram Methods 0.000 description 3
- 238000013473 artificial intelligence Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 2
- 235000013399 edible fruits Nutrition 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000008901 benefit Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 238000001914 filtration Methods 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000002401 inhibitory effect Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000012545 processing Methods 0.000 description 1
- 238000011160 research Methods 0.000 description 1
- 238000012827 research and development Methods 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000000717 retained effect Effects 0.000 description 1
Abstract
The invention discloses a kind of improved object detection method, system and device, wherein method is the following steps are included: obtain original picture to be detected, and after dividing according to the first predetermined manner to original picture, obtains n region pictures；After successively carrying out Multi-layer technology to each region picture, n group first partial picture is obtained；After the size of first partial picture is amplified to the size of original picture, the local picture of n group second is obtained；After carrying out target detection to the second local picture using preset detection model, one second local picture is obtained from the local picture of each group second respectively according to testing result；The local picture of n second is fused into a detection picture according to the second predetermined manner, and after will test the size reduction to the size of original picture of picture, output detects picture.The present invention effectively reduces algorithm of target detection to the probability of Small object missing inspection and false retrieval, greatly improves the accuracy of detection, can be widely applied to detection technique field.
Description
Technical field
The present invention relates to detection technique field more particularly to a kind of improved object detection methods, system and device.
Background technique
Into after 21 century, with the fast development of computer science, artificial intelligence the relevant technologies have the variation of matter, deep
Degree study is one of the priority research areas of current artificial intelligence, and the algorithm of target detection based on deep learning is that today's society is ground
The important topic studied carefully.For the task of target detection, researcher mainly pursues two targets: first is that accuracy rate is high, second is that
Speed is fast, this is also two important indicators for measuring detection algorithm quality.
Deep learning is also traced back education sector direction is unrestrained among generally the combining of theory and practice.It is deep in recent years
Degree study is quickly grown in education sector, and expenditure of the China in terms of education also greatly improves, so being based on deep learning
Target detection research and development seem most important in education sector.China wants to develop education, and must just follow the epoch
Paces are realized for student's action recognition under classroom environment, and intelligence of the deep learning in terms of education is embodied.Currently based on
The Human bodys' response method of deep learning algorithm has very much, such as SSD, YOLO algorithm based on homing method, is based on region side
The Fast RCNN of method, R FCN algorithm etc..These detection algorithms are mostly input picture to be not added any pretreatment, directly defeated
Enter and carry out feature extraction in convolutional neural networks, is divided by convolution, pond sequence of operations and then using classifier
Class finally filters out optimal testing result figure using NMS algorithm.Above-mentioned some algorithm of target detection, although meeting reality
The requirement of when property, but to the detection accuracy of some Small objects Small object of movement (such as classroom environment Students ') but it is insufficient,
The case where recall rate is low, there is also missing inspection false retrieval sometimes.Since the target detection based on recurrence uses multi-scale method,
Convolution receptive field on different scale is different, especially high-level convolutional layer, and receptive field is also very big.Therefore, for high level
Grade characteristic layer, feature extraction content is more abstract, and feature extraction is more abstract, and corresponding detailed information is fewer, thus to small
The detection of target is insensitive, results in lower to Small object object detection precision.
Summary of the invention
In order to solve the above-mentioned technical problem, the object of the present invention is to provide a kind of pair of more sensitive target detections of Small object
Method, system and device.
Technical solution used by the method for the present invention is:
A kind of improved object detection method, comprising the following steps:
Original picture to be detected is obtained, and after dividing according to the first predetermined manner to original picture, obtains n administrative division maps
Piece；
After successively carrying out Multi-layer technology to each region picture, n group first partial picture, every group of first partial picture packet are obtained
Include m first partial pictures；
After the size of first partial picture is amplified to the size of original picture, the local picture of n group second is obtained；
After carrying out target detection to the second local picture using preset detection model, according to testing result respectively from each group
One second local picture is obtained in second local picture, to obtain n second local picture；
N second local picture is fused into a detection picture according to the second predetermined manner, and will test the ruler of picture
It is very little be contracted to the size of original picture after, output detection picture.
Further, described to obtain original picture to be detected, and after being divided according to the first predetermined manner to original picture, it obtains
The step for obtaining n region pictures, specifically:
Obtain original picture to be detected, and original picture is fifty-fifty divided into top left region, right regions, lower left region,
After lower right area and central region, five region pictures are obtained.
Further, it is described Multi-layer technology successively is carried out to each region picture after, obtain n group first partial picture, every group the
The step for one local picture includes m first partial pictures, specifically:
After successively carrying out the Multi-layer technology that five times have repeating part to each region picture, five groups of first partial pictures are obtained,
Every group of first partial picture includes five first partial pictures.
Further, it is described target detection is carried out to the second local picture using preset detection model after, tied according to detection
Fruit obtains one second local picture from the local picture of each group second respectively, to obtain the step for n opens the second local picture,
Specifically:
The algorithm of target detection based on convolutional neural networks is used to carry out feature extraction to the second local picture and detect；
Screening Treatment is carried out to the second local picture after testing in conjunction with testing result and preset inhibition screening technique
Afterwards, an optimal second local picture is obtained, from the local picture of each group second respectively to obtain five second local pictures.
Further, described that n second local picture is fused into a detection picture according to the second predetermined manner, and will inspection
After the size reduction of mapping piece to the size of original picture, the step for detecting picture is exported, specifically:
Five second local pictures are spliced according to preset position and are fused into a detection picture, and will test
After the size reduction of picture to the size of original picture, output detection picture.
Further, the central region is a rectangle, and the central region is obtained in the following manner:
Obtain central point of the central point of original picture as rectangle；
Obtain length of the half as rectangle of the length of original picture, and acquisition original picture width half as square
The width of shape；
After establishing rectangle in conjunction with the central point of rectangle, length and width, using rectangle as central region.
Further, the central region is quadrangle, and the central region is obtained in the following manner:
Four midpoints are sequentially connected to obtain the first quadrangle behind the midpoint of four edges in acquisition original picture respectively；
The area of first quadrangle is scaled to after a quarter, obtaining the second quadrangle as central region,
The central point of second quadrangle is overlapped with the central point of original picture.
Technical solution used by present system is:
A kind of improved object detection system, comprising:
Division module, for obtaining original picture to be detected, and after being divided according to the first predetermined manner to original picture,
Obtain n region pictures；
Hierarchical block, after successively carrying out Multi-layer technology to each region picture, acquisition n group first partial picture, every group
First partial picture includes m first partial pictures；
Amplification module obtains the second part of n group after the size of first partial picture is amplified to the size of original picture
Picture；
Detection module, after carrying out target detection to the second local picture using preset detection model, according to detection
As a result one second local picture is obtained, from the local picture of each group second respectively to obtain n second local picture；
Fusion Module, for n second local picture to be fused into a detection picture according to the second predetermined manner, and will
After the size reduction to the size of original picture for detecting picture, output detection picture.
Further, the division module is specifically used for obtaining original picture to be detected, and original picture is fifty-fifty divided into
After top left region, right regions, lower left region, lower right area and central region, five region pictures are obtained.
Technical solution used by apparatus of the present invention is:
A kind of improved object detecting device, comprising:
At least one processor；
At least one processor, for storing at least one program；
When at least one described program is executed by least one described processor, so that at least one described processor is realized
A kind of above-mentioned improved object detection method.
The beneficial effects of the present invention are: the present invention obtains more offices by dividing to original picture and Multi-layer technology
Portion's figure layer, and each local figure layer is detected, increase the attention rate of regional area, keeps the detection accuracy to local Small object big
It is big to improve, solve the problems, such as it is insensitive to small target deteection, effectively reduce algorithm of target detection to Small object missing inspection and
The probability of false retrieval greatly improves the accuracy of detection.
Detailed description of the invention
Fig. 1 is a kind of step flow chart of improved object detection method of the present invention；
Fig. 2 is a kind of structural block diagram of improved object detection system of the present invention；
Fig. 3 is a kind of structural schematic diagram of embodiment of central region in specific embodiment；
Fig. 4 is the structural schematic diagram of the another embodiment of central region in specific embodiment.
Specific embodiment
Embodiment one
As shown in Figure 1, present embodiments providing a kind of improved object detection method, comprising the following steps:
S1, original picture to be detected is obtained, and after dividing according to the first predetermined manner to original picture, obtains n areas
Domain picture.
S2, after successively carrying out Multi-layer technology to each region picture, n group first partial picture, every group of first partial figure are obtained
Piece includes m first partial pictures.
S3, after the size of first partial picture is amplified to the size of original picture, the local picture of n group second is obtained.
S4, after carrying out target detection to the second local picture using preset detection model, according to testing result respectively from
One second local picture is obtained in the local picture of each group second, to obtain n second local picture.
S5, n second local picture is fused into a detection picture according to the second predetermined manner, and will test picture
After size reduction to the size of original picture, output detection picture.
The working principle of the above method are as follows: for the picture of an arbitrary size, before needing detection, first to whole picture into
The operation of two steps of row.First, the division in equal regions is carried out to picture, is divided into the region of n equal sizes, is referred to as respectively
For region one, region two, up to region n, the n can be arranged according to distribution of the detection target in figure, if central area
There is conspicuous object, then 5 is chosen, if center picture can choose 4 without conspicuous object.Second, to the picture in this five regions again into
The extraction of row component layer will extract the component layer Chong Die with the m of the sizes such as region picture for the picture in each region,
The m is the integer more than or equal to 2, can increase the size of m value according to figure Small Target quantity.Local picture in every group has
The part of overlapping, it is also possible to which the local picture in different groups has a part of overlapping, finally obtained to be n*m and have overlay region
The local picture in domain, then the size of this n*m picture is amplified to size identical with input original picture, obtain the second Local map
Piece.Second local picture conveying is such as preset in trained detection model, carries out target detection, and according to the knot of target detection
Fruit filters out an optimal second local picture from the local picture of each group second, to obtain n second local picture.It will sieve
The n selected second local picture is merged, and one and the consistent full picture of original image content are combined into.It finally again will figure
Piece is contracted to the size of original image, until the final result of output detection.In the above method, to the picture of input before detection into
Original picture is carried out division and Multi-layer technology, obtains more subgraph layers by row pretreatment, after being amplified to local figure layer,
Target detection is carried out, increases the attention rate of regional area, greatly improves the detection accuracy to local Small object, solve to small
The insensitive problem of target detection effectively reduces algorithm of target detection to the probability of Small object missing inspection and false retrieval, greatly
Improve the accuracy of detection.
Wherein, step S1 specifically: obtain original picture to be detected, and original picture is fifty-fifty divided into top left region,
After right regions, lower left region, lower right area and central region, five region pictures are obtained.
When dividing to original picture, more parts can be divided into, for example is symmetrically divided into 4 parts, it can also be according to detection mesh
It marks present center and is divided into 5 parts, original picture is divided into 5 parts, respectively top left region, upper right in the present embodiment
Region, lower left region, lower right area and central region, the central region respectively with top left region, right regions, lower left region
There is the part of coincidence with lower right area, four top left region, right regions, lower left region and lower right area regions can spell
It is connected into the original picture of a completion.This is because the effective information of the central region of picture is more, therefore added on four regions
The central point of central region, central region is overlapped with the central point of input picture, and the central region can use a variety of realizations
Mode, in the present embodiment, a kind of embodiment provided are as follows:
Referring to Fig. 3, the central region is a rectangle, and the central point of the rectangle is overlapped with the central point of original picture, institute
The half of the length of a length of original picture of rectangle is stated, the width of the rectangle is the wide half of original picture, and the rectangle is middle part
Region.
It is as follows to the another embodiment of central region offer in the present embodiment:
Referring to Fig. 4, the central region is quadrangle, and four points of the quadrangle are issued respectively in the symmetrical of original picture
On axis, specifically, this four points are respectively the midpoint of the line at the central point of original picture and the midpoint of four edges, the quadrangle
Central point is overlapped with the central point of original picture.When being sequentially connected the midpoint on four side of original picture, a big quadrangle is obtained, it will
The big quadrangle according to after the scale smaller of side length 1:2 to get arrive the quadrangle, using the quadrangle as central region.
The method of both above-mentioned selection central regions is provided to be truncated when avoiding detection target by region division
The case where, it avoids as far as possible, larger target is truncated, and improves detection effect.
Wherein, the step S2, by referring to the specific steps for Fig. 3 are as follows: successively each region picture have for five times
After the Multi-layer technology of repeating part, five groups of first partial pictures are obtained, every group of first partial picture includes five first partial figures
Piece.
After getting region picture, region ready-portioned for above five, which carries out 5 width, has the picture of repeating part to extract,
Withdrawal ratio accounts for 4/5 region of original picture, and the range of the entire picture of span in extraction process, degree of overlapping is moderate, need to be all packets
Extracted region containing target comes out., picture identical as original picture size is obtained in this way accounts for the 25 width Local maps that original image ratio is 1/5
Piece, it is therefore an objective to which " wisp " on original picture is become into " the big object " for target detection.
Wherein, the step S4, specifically includes step S41~S42:
S41, it uses the algorithm of target detection based on convolutional neural networks feature extraction is carried out simultaneously to the second local picture
Detection.
S42, the second local picture after testing is screened in conjunction with testing result and preset inhibition screening technique
After processing, an optimal second local picture is obtained, from the local picture of each group second respectively to obtain five second parts
Picture.
Ready-portioned figure layer is input in the detection model based on convolutional neural networks, feature extraction and classifying is carried out,
Achieve the effect that target detection, contain database in the detection model, which is homemade students ' behavior movement number
According to library, the model that database is trained by the algorithm of convolutional neural networks and is needed.To all local pictures
After carrying out target detection, inhibits filtering algorithm to delete and screen local picture using component layer, effectively delete component
Most of overlappings in layer (i.e. second local picture) testing result and the target frame being truncated, retain correct targets more as far as possible
Frame, then compared by the detection block of component layer and original image layer, degree of overlapping is highest to be retained, remaining is suppressed, and obtains five
A optimal component layer.
Wherein, step S5 specifically: five second local pictures are spliced according to preset position and are fused into one
Detection picture, and after will test the size reduction to the size of original picture of picture, output detection picture.
Original image is cut into five regions by region division, and each piece of region is the 1/4 of original image, when detecting puts region
The big size for arriving original picture.5 optimal component layers can be obtained after inhibiting screening by figure layer, region fusion is carried out to it, is made
Be stitched together, obtain final result figure.
The above method carries out input picture to carry out target detection after region division and figure layer extract, then after will test
Local picture carries out fusion treatment, effectively increases the attention rate of regional area, makes to the detection accuracy of local Small object significantly
It improves, solves the disadvantages such as most algorithm of target detection are not high to the detection accuracy of Small object, recall rate is low, effectively reduce
Algorithm of target detection greatly improves the accuracy of detection to the probability of Small object missing inspection and false retrieval.
In addition, this method has structure simple, easy to accomplish, the advantage easily combined with deep learning algorithm is designed, not only
It can be applied to the Small object person detecting under classroom environment, can also be applied to more due to Small object object detection precision
Under not high any scene.
Embodiment two
As shown in Fig. 2, present embodiments providing a kind of improved object detection system, comprising:
Division module, for obtaining original picture to be detected, and after being divided according to the first predetermined manner to original picture,
Obtain n region pictures；
Hierarchical block, after successively carrying out Multi-layer technology to each region picture, acquisition n group first partial picture, every group
First partial picture includes m first partial pictures；
Amplification module obtains the second part of n group after the size of first partial picture is amplified to the size of original picture
Picture；
Detection module, after carrying out target detection to the second local picture using preset detection model, according to detection
As a result one second local picture is obtained, from the local picture of each group second respectively to obtain n second local picture；
Fusion Module, for n second local picture to be fused into a detection picture according to the second predetermined manner, and will
After the size reduction to the size of original picture for detecting picture, output detection picture.
It is further used as preferred embodiment, the division module is specifically used for obtaining original picture to be detected, and will
After original picture is fifty-fifty divided into top left region, right regions, lower left region, lower right area and central region, five areas are obtained
Domain picture.
Above system obtains more local figure layers, and to each part by carrying out division and Multi-layer technology to original picture
Figure layer is detected, and the attention rate of regional area is increased, and greatly improves the detection accuracy to local Small object, solves to small
The insensitive problem of target detection effectively reduces algorithm of target detection to the probability of Small object missing inspection and false retrieval, greatly
Improve the accuracy of detection.
Embodiment three
Present embodiments provide a kind of improved object detecting device, comprising:
At least one processor；
At least one processor, for storing at least one program；
When at least one described program is executed by least one described processor, so that at least one described processor is realized
A kind of improved object detection method described in embodiment one.
One kind provided by embodiment of the present invention method one can be performed in a kind of improved object detecting device of the present embodiment
Improved object detection method, any combination implementation steps of executing method embodiment, have the corresponding function of this method and
Beneficial effect.
It is to be illustrated to preferable implementation of the invention, but the invention is not limited to the implementation above
Example, those skilled in the art can also make various equivalent variations on the premise of without prejudice to spirit of the invention or replace
It changes, these equivalent deformations or replacement are all included in the scope defined by the claims of the present application.
Claims (10)
1. a kind of improved object detection method, which comprises the following steps:
Original picture to be detected is obtained, and after dividing according to the first predetermined manner to original picture, obtains n region pictures；
After successively carrying out Multi-layer technology to each region picture, n group first partial picture is obtained, every group of first partial picture includes m
Open first partial picture；
After the size of first partial picture is amplified to the size of original picture, the local picture of n group second is obtained；
After carrying out target detection to the second local picture using preset detection model, according to testing result respectively from each group second
One second local picture is obtained in local picture, to obtain n second local picture；
N second local picture is fused into a detection picture according to the second predetermined manner, and will test the size contracting of picture
As low as after the size of original picture, output detection picture.
2. a kind of improved object detection method according to claim 1, which is characterized in that described to obtain original to be detected
Picture, and after being divided according to the first predetermined manner to original picture, the step for obtaining n region pictures, specifically:
Original picture to be detected is obtained, and original picture is fifty-fifty divided into top left region, right regions, lower left region, bottom right
Behind region and central region, five region pictures are obtained.
3. a kind of improved object detection method according to claim 2, which is characterized in that described successively to each administrative division map
Piece carry out Multi-layer technology after, obtain n group first partial picture, every group of first partial picture include m open first partial pictures this
Step, specifically:
After successively carrying out the Multi-layer technology that five times have repeating part to each region picture, five groups of first partial pictures of acquisition, every group
First partial picture includes five first partial pictures.
4. a kind of improved object detection method according to claim 3, which is characterized in that described to use preset detection
After model carries out target detection to the second local picture, one is obtained from the local picture of each group second respectively according to testing result
Second local picture, to obtain the step for n opens the second local picture, specifically:
The algorithm of target detection based on convolutional neural networks is used to carry out feature extraction to the second local picture and detect；
After carrying out Screening Treatment to the second local picture after testing in conjunction with testing result and preset inhibition screening technique, point
An optimal second local picture is not obtained, from the local picture of each group second to obtain five second local pictures.
5. a kind of improved object detection method according to claim 4, which is characterized in that described by n second part
Picture is fused into a detection picture according to the second predetermined manner, and will test the size reduction of picture to the size of original picture
Afterwards, the step for output detection picture, specifically:
Five second local pictures are spliced according to preset position and are fused into a detection picture, and will test picture
Size reduction to the size of original picture after, output detection picture.
6. a kind of improved object detection method according to claim 2, which is characterized in that the central region is a square
Shape, the central region are obtained in the following manner:
Obtain central point of the central point of original picture as rectangle；
Obtain length of the half as rectangle of the length of original picture, and acquisition original picture width half as rectangle
Width；
After establishing rectangle in conjunction with the central point of rectangle, length and width, using rectangle as central region.
7. a kind of improved object detection method according to claim 2, which is characterized in that the central region is four sides
Shape, the central region are obtained in the following manner:
Four midpoints are sequentially connected to obtain the first quadrangle behind the midpoint of four edges in acquisition original picture respectively；
The area of first quadrangle is scaled to after a quarter, obtaining the second quadrangle as central region, it is described
The central point of second quadrangle is overlapped with the central point of original picture.
8. a kind of improved object detection system characterized by comprising
Division module is obtained for obtaining original picture to be detected, and after dividing according to the first predetermined manner to original picture
N region pictures；
Hierarchical block, after successively carrying out Multi-layer technology to each region picture, acquisition n group first partial picture, every group first
Local picture includes m first partial pictures；
Amplification module obtains the second Local map of n group after the size of first partial picture is amplified to the size of original picture
Piece；
Detection module, after carrying out target detection to the second local picture using preset detection model, according to testing result
One second local picture is obtained, from the local picture of each group second respectively to obtain n second local picture；
Fusion Module for n second local picture to be fused into a detection picture according to the second predetermined manner, and will test
After the size reduction of picture to the size of original picture, output detection picture.
9. a kind of improved object detection system according to claim 8, which is characterized in that the division module is specifically used
In acquisition original picture to be detected, and original picture is fifty-fifty divided into top left region, right regions, lower left region, bottom right area
Behind domain and central region, five region pictures are obtained.
10. a kind of improved object detecting device characterized by comprising
At least one processor；
At least one processor, for storing at least one program；
When at least one described program is executed by least one described processor, so that at least one described processor realizes right
It is required that a kind of described in any item improved object detection methods of 1-7.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811652762.4A CN109741333A (en) | 2018-12-28 | 2018-12-28 | A kind of improved object detection method, system and device |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811652762.4A CN109741333A (en) | 2018-12-28 | 2018-12-28 | A kind of improved object detection method, system and device |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109741333A true CN109741333A (en) | 2019-05-10 |
Family
ID=66362918
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811652762.4A Pending CN109741333A (en) | 2018-12-28 | 2018-12-28 | A kind of improved object detection method, system and device |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109741333A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112565614A (en) * | 2021-02-22 | 2021-03-26 | 四川赛狄信息技术股份公司 | Signal processing module and method |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108446694A (en) * | 2017-02-16 | 2018-08-24 | 杭州海康威视数字技术股份有限公司 | A kind of object detection method and device |
| CN109033950A (en) * | 2018-06-12 | 2018-12-18 | 浙江工业大学 | Vehicle based on multiple features fusion cascade deep model, which is disobeyed, stops detection method |
-
2018
- 2018-12-28 CN CN201811652762.4A patent/CN109741333A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108446694A (en) * | 2017-02-16 | 2018-08-24 | 杭州海康威视数字技术股份有限公司 | A kind of object detection method and device |
| CN109033950A (en) * | 2018-06-12 | 2018-12-18 | 浙江工业大学 | Vehicle based on multiple features fusion cascade deep model, which is disobeyed, stops detection method |
Non-Patent Citations (2)
| Title |
|---|
| YONGXI LU 等: "Adaptive Object Detection Using Adjacency and Zoom Prediction", 《ARXIV》 * |
| 吴亚熙 等: "基于卷积神经网络的多层级目标检测方法", 《软件》 * |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112565614A (en) * | 2021-02-22 | 2021-03-26 | 四川赛狄信息技术股份公司 | Signal processing module and method |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108898047B (en) | Pedestrian detection method and system based on blocking and shielding perception | |
| CN106683091A (en) | Target classification and attitude detection method based on depth convolution neural network | |
| CN110263705A (en) | Towards two phase of remote sensing technology field high-resolution remote sensing image change detecting method | |
| Henriques et al. | The role of the lithological setting on the landslide pattern and distribution | |
| CN107133955B (en) | A kind of collaboration conspicuousness detection method combined at many levels | |
| CN109711288A (en) | Remote sensing ship detecting method based on feature pyramid and distance restraint FCN | |
| CN110490202A (en) | Detection model training method, device, computer equipment and storage medium | |
| CN107315755A (en) | The orbit generation method and device of query object | |
| CN104182985B (en) | Remote sensing image change detection method | |
| CN109886245A (en) | A kind of pedestrian detection recognition methods based on deep learning cascade neural network | |
| CN110414380A (en) | A kind of students ' behavior detection method based on target detection | |
| CN106650600A (en) | Forest smoke and fire detection method based on video image analysis | |
| CN107507170A (en) | A kind of airfield runway crack detection method based on multi-scale image information fusion | |
| CN105608454A (en) | Text structure part detection neural network based text detection method and system | |
| CN103761526B (en) | Urban area detecting method based on feature position optimization and integration | |
| CN109657602A (en) | Automatic functional region of city method and system based on streetscape data and transfer learning | |
| CN107292886A (en) | Object intrusion detection method and device based on mesh generation and neutral net | |
| CN109740676A (en) | Object detection moving method based on similar purpose | |
| CN109241902A (en) | A kind of landslide detection method based on multi-scale feature fusion | |
| CN109657622A (en) | A kind of detection of traffic lights and recognition methods, device and equipment | |
| CN108776777A (en) | The recognition methods of spatial relationship between a kind of remote sensing image object based on Faster RCNN | |
| Xia et al. | Deep extraction of cropland parcels from very high-resolution remotely sensed imagery | |
| Zhang et al. | Detecting fractional land-cover change in arid and semiarid urban landscapes with multitemporal Landsat Thematic mapper imagery | |
| CN107977667A (en) | SAR target discrimination methods based on semi-supervised coorinated training | |
| CN114140665A (en) | Dense small target detection method based on improved YOLOv5 |
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 |
Application publication date: 20190510 |
|
| RJ01 | Rejection of invention patent application after publication |