CN109858493A - A kind of dimension self-adaption nuclear phase pass neural network based filter tracking method - Google Patents
A kind of dimension self-adaption nuclear phase pass neural network based filter tracking method Download PDFInfo
- Publication number
- CN109858493A CN109858493A CN201811619242.3A CN201811619242A CN109858493A CN 109858493 A CN109858493 A CN 109858493A CN 201811619242 A CN201811619242 A CN 201811619242A CN 109858493 A CN109858493 A CN 109858493A
- Authority
- CN
- China
- Prior art keywords
- target
- filter
- target area
- neural network
- characteristic pattern
- 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
Landscapes
- Image Analysis (AREA)
Abstract
The present invention provides a kind of dimension self-adaption nuclear phase pass neural network based filter tracking method, and method includes: to obtain initial frame data, target area and target area parameter information；At least two layers of feature of target area are extracted as characteristic pattern；Extracted characteristic pattern is transformed into continuous domain；By minimizing the loss function of filter response output and desired output, effective convolution filter is calculated, and target prodiction is carried out based on effective convolution filter；Dimensional variation based on scaling filter prediction target；Whether the peak response output valve for cascading the classification output and effective convolution filter of multiple target classifier, judge target with losing.Using the embodiment of the present invention, target area convolution feature is extracted as target signature using lightweight neural network MobileNetV2 on the basis of useful space convolution filter, enhances filter robustness, reduces model parameter quantity, promote tracking velocity.
Description
Technical field
The present invention relates to filter tracking technical fields, more particularly to a kind of dimension self-adaption nuclear phase neural network based
Close filter tracking method.
Background technique
Target tracking algorism obtained great development in recent years, was widely used in video monitoring, robotic tracking, man-machine
The fields such as interactive and intelligent transportation.The combination of deep learning and correlation filtering greatly promotes the performance of tracker, but by
In the network of the very big limit algorithm tracking velocity of the complexity of network, lightweight the same of tracking accuracy can not be being lost substantially
When, the calculating speed of very big track algorithm.Simultaneously one fast and accurately size measurement strategy also can more preferably guarantee to track
The accurately and quickly property of process.
Track algorithm of the conventional depth study in conjunction with correlation filtering is all trained using very big basic network, such as
VGGNet, GoogleNet etc..Since above-mentioned model parameter is excessive, generally takes on size measurement and existed using translation filter
The dimensional variation of prediction target is repeated several times in the target of different resolution, can greatly limit the precision and speed of tracking in this way.
Summary of the invention
In view of the foregoing deficiencies of prior art, the purpose of the present invention is to provide a kind of scales neural network based
Self-adaptive kernel correlation filtering tracking, it is intended to improve the precision and speed for carrying out target predicting and tracking.
In order to achieve the above objects and other related objects, the present invention provides a kind of dimension self-adaption core neural network based
Correlation filtering tracking, which comprises
Obtain initial frame data, target area and the target area parameter information, wherein the target area parameter
Information includes at least a coordinate points and size for the target area；
At least two layers of feature of the target area are extracted as characteristic pattern；
Extracted characteristic pattern is transformed into continuous domain；
By minimizing the loss function of filter response output and desired output, effective convolution filter, and base are calculated
Target prodiction is carried out in effective convolution filter；
Dimensional variation based on scaling filter prediction target；
According to the target prodiction and the target scale predicted variation, the target area of prediction is obtained, filter is found out
Wave device determines the affiliated example classification in the target area in the target area peak response output valve, and using classifier, connection
It closes filter and judges target whether with losing in the output of the region peak response output valve and the classifier.
In a kind of implementation of the invention, the associated filters are in the region peak response output valve and the classification
The output of device judges the step of whether target is with losing, comprising:
Based on one multi-class classifier of MobileNetV2 network training, calculating and exporting target is each classification
Probability；
The maximum value is determined as target detection classification by the maximum value for determining probability；
Judge whether the peak response output valve of effective convolution filter is greater than threshold value；
If it is, indicating that target is not lost；
If not, judging whether the target detection classification is consistent with target concrete class again；
If be consistent, then it is assumed that the target is not lost, and thinks that the target is lost if not being consistent.
In a kind of implementation of the invention, it is described based on scaling filter prediction target dimensional variation the step of, packet
It includes:
Calculate the operation result of scaling filter and training sample characteristic pattern；
Calculate the mean square error with desired output；
Mean square error by minimizing desired output and calculated result acquires optimal scale filter, acquires scale filter
The response of device exports.
In a kind of implementation of the invention, at least two layers of feature for extracting the target area are as characteristic pattern
The step of, comprising:
Determine MobileNetV2 network inputs size；
Extract 24 conv2/1/sum layers of characteristic pattern of dimension, the 96 conv5/2/sum layers of characteristic pattern of dimension of the target area.
In a kind of implementation of the invention, described the step of extracted characteristic pattern is transformed into continuous domain, comprising:
Using interpolation operation, extracted characteristic pattern is transformed into continuous domain.
In a kind of implementation of the invention, the corresponding function of the interpolation operation is embodied are as follows:
Wherein, S (γ) is interpolating function, and a is settable constant, and γ is the position of interpolation point and pixel.
As described above, a kind of dimension self-adaption nuclear phase neural network based provided in an embodiment of the present invention closes filter tracking
Method, in useful space convolution filter (Efficient Convolution Operators for Tracking, ECO)
On the basis of using lightweight neural network MobileNetV2 extract target area convolution feature be used as target signature, enhancing filter
Wave device robustness reduces model parameter quantity, promotes tracking velocity；Additional unidimensional scale self-adaptive kernel correlation filter is added,
The individually training, local optimum of two filters, can precisely quick predict target scale variation；The multi-class classifier of training, grade
Join correlation filtering response output, precisely detects whether target loses.Experiment shows the present invention in target scale variation, deformation, screening
There is good adaptability under the complicated tracking scene such as gear.
Detailed description of the invention
Fig. 1 is that a kind of dimension self-adaption nuclear phase neural network based of the embodiment of the present invention closes the one of filter tracking method
Kind flow diagram.
Fig. 2 is that a kind of dimension self-adaption nuclear phase neural network based of the embodiment of the present invention closes the another of filter tracking method
A kind of flow diagram.
Fig. 3 is that a kind of dimension self-adaption nuclear phase neural network based of the embodiment of the present invention closes the one of filter tracking method
Kind effect diagram；
Fig. 4 is that a kind of dimension self-adaption nuclear phase neural network based of the embodiment of the present invention closes the another of filter tracking method
A kind of effect diagram；
Fig. 5 is that a kind of dimension self-adaption nuclear phase neural network based of the embodiment of the present invention closes filter tracking method again
A kind of effect diagram；
Fig. 6 is the tracking effect schematic diagram of the prior art.
Specific embodiment
Illustrate embodiments of the present invention below by way of specific specific example, those skilled in the art can be by this specification
Other advantages and efficacy of the present invention can be easily understood for disclosed content.The present invention can also pass through in addition different specific realities
The mode of applying is embodied or practiced, the various details in this specification can also based on different viewpoints and application, without departing from
Various modifications or alterations are carried out under spirit of the invention.
Please refer to Fig. 1-5.It should be noted that only the invention is illustrated in a schematic way for diagram provided in the present embodiment
Basic conception, only shown in schema then with related component in the present invention rather than component count, shape when according to actual implementation
Shape and size are drawn, when actual implementation kenel, quantity and the ratio of each component can arbitrarily change for one kind, and its component cloth
Office's kenel may also be increasingly complex.
As depicted in figs. 1 and 2, the embodiment of the present invention provides a kind of dimension self-adaption core correlation filtering neural network based
Tracking, which comprises
S101 obtains initial frame data, target area and the target area parameter information, wherein the target area
Field parameter information includes at least a coordinate points and size for the target area.
Read initial frame data and target area, target area parameter information, illustratively, target area parameter information
Including target area top left co-ordinate (x, y), target wide w and high h.
S102 extracts at least two layers of feature of the target area as characteristic pattern.
Target area conv2/1/sum layers and conv5/2/sum layers is extracted using MobileNetV2 convolutional neural networks
Feature is as clarification of objective figure, illustratively, for the characteristic pattern for the D dimension that j-th of training sample includes are as follows:Wherein, x represents training sample, and subscript D represents each training sample characteristic dimension as D dimension.
Illustratively, MobileNetV2 network inputs size is 300 × 300, extracts target area 24 and ties up conv2/1/
Sum layers totally 120 dimensional features are as clarification of objective figure with 96 conv5/2/sum layers of dimensions, so D value of the present invention is 120.
Extracted characteristic pattern is transformed into continuous domain by S103.
In a kind of implementation of the invention, described the step of extracted characteristic pattern is transformed into continuous domain, comprising: benefit
With interpolation operation, extracted characteristic pattern is transformed into continuous domain.
Characteristic pattern is transformed into continuous domain J using interpolation operationd{xd(t), conversion formula is as follows:
D indicates d dimensional feature figure, NdRepresent characteristic patternSpace pixel number, T is arbitrary constant, indicates coordinate
System's scaling, bdInterpolation operation is represented, x indicates training sample.
During specifically, the corresponding function of the interpolation operation is embodied are as follows:
Wherein, S (γ) is interpolating function, and a is settable constant, and value of the present invention is that 0.5, γ is interpolation point and pixel
The position of point.
In the embodiment of the present invention, trace model is established using effective convolution filter, is prediction target translation position.Such as figure
Shown in 3,1 is a convolutional layer, and 2 be a convolutional layer, and 3 be that convolutional layer corresponding to convolutional layer and 2 corresponding to 1 carries out continuously
Treated in domain comparison diagram.
S104 responds the loss function of output with desired output by minimizing filter, calculates effective convolution filter,
And target prodiction is carried out based on effective convolution filter.
Specifically, by solving detection output SpfThe loss function of { x } and desired output y can be filtered in the hope of active volume product
Wave device f.Loss function E (f, P) is shown below:
Wherein, M indicates that a shared M training sample, w are space regular terms, ajIndicate that weight coefficient, λ indicate regular terms
Coefficient, P are D × C matrix, and C filter linearity used in the corresponding filter of each dimensional characteristics is combined in every a line expression
Coefficient, detection output Spf{ x } expression formula are as follows:
Wherein, * operator representation space related operation accords with, it is assumed that regularization coefficient λ value is 0.0075, space regular terms
W and desired output y is the output of dimensional Gaussian shape, output function are as follows:
Illustratively, region 4 as shown in Figure 3 is target following region.
S105, the dimensional variation based on scaling filter prediction target.
In the embodiment of the present invention, by adding unidimensional scale self-adaptive kernel correlation filter model, the scale of target is predicted
Variation.The modeling of scaling filter is by calculating scaling filter hsWith the correlation result of training sample x, then calculate
With desired output gsMean square error, finally by mean square error minimum acquire optimal scale filter hs, it is shown below:
Above formula λsIotazation constant is represented, K represents characteristic dimension, and subscript d represents d dimensional feature, amounts to K dimension.It is logical
H can be obtained with rapid solving by crossing Paasche Wa Er theorem and being transferred to frequency domainsIn the expression formula of frequency domainAre as follows:
To scaling filter h in the way of linear weighted functionsIt is updated, is shown below:
At,BtIt respectively representsThe molecule of expression formula, denominator part, η represent learning rate, following table t, and t-1 is respectively represented
Present frame and former frame.Scaling filter h can finally be acquiredsFrequency domain response export YsAre as follows:
Wherein, I represents present frame input.Scale response output Y is obtained by Fourier inversionsIn spatial domain ysTable
It is shown below up to formula:
The final target scale that divides is s scale, finds out space and response output ysPosition where maximum value is usedTable
Show,Target prediction scale size is anw×anH, wherein a represents scale factor, and w, h are represented
The width and height of target area.
Wherein, learning rate η value is 0.001, iotazation constant λsValue is 0.0075, spends filter hsUsed
Feature is Fast Field histogram feature (Fast Histogram of Oriented Gradient, FHOG) feature of 31 dimensions
Dimension, characteristic dimension K value are 31, desired output gsValue is the output of one-dimensional gaussian shape, and scale factor a is 1.02.
S106 changes according to the target prodiction and the target scale predicted, obtains the target area of prediction, look for
Filter determines the affiliated example class in the target area in the target area peak response output valve, and using classifier out
Not, whether associated filters judge target with losing in the output of the region peak response output valve and the classifier.
In the prior art, whether most of algorithms only judge target with losing with single threshold method, and judging result accuracy is not high,
Specific implementation are as follows: one threshold value T of setting judges that peak response exports YmaxJudge with the relationship of threshold value T target whether with losing,
It is shown below:
In a kind of implementation of the invention, the associated filters are in the region peak response output valve and the classification
The output of device judges the step of whether target is with losing, comprising:
Based on one multi-class classifier of MobileNetV2 network training, calculating and exporting target is each classification
Probability；
The maximum value is determined as target detection classification by the maximum value for determining probability；
Judge the peak response output valve Y of effective convolution filtermaxWhether threshold value T is greater than；
If it is, indicating that target is not lost；
If not, judging whether the target detection classification is consistent with target concrete class again；
If be consistent, then it is assumed that the target is not lost, and thinks that the target is lost if not being consistent.
The present invention is based on one multi-class classifiers of MobileNetV2 network training, and exporting target is each classification
Probability, it is assumed that are as follows: (P1,P2,, Pn), choose wherein maximum probability PmaxFor the detection classification C of the targetp.Compare filtering first
Device peak response output valve YmaxWhether it is greater than threshold value T, then thinks that target is not lost if it is greater than threshold value T, if it is less than threshold value
T, then judge target detection classification CpWith target concrete class CtWhether it is consistent, is consistent, thinks that the target is not lost, be not consistent
Then think that the target is lost.It wherein, is 21 classes based on the polytypic classification of MobileNetV2 network training, threshold value T value is
0.3。
Using the embodiment of the present invention, qualitative analysis effect is as shown in figure 4, compared with existing monodrome comparison method, the present invention
The effect of the actual treatment of embodiment is as shown in figure 5, the target following using the prior art is as shown in Figure 6.
Therefore, using the embodiment of the present invention, in useful space convolution filter (Efficient Convolution
Operators for Tracking, ECO) on the basis of, target area is extracted using lightweight neural network MobileNetV2
Domain convolution feature enhances filter robustness as target signature, reduces model parameter quantity, promotes tracking velocity；Addition
Additional unidimensional scale self-adaptive kernel correlation filter, the individually training, local optimum of two filters can accurate quick predict mesh
Scale variation；The multi-class classifier of training, cascade correlation filtering response output, precisely detects whether target loses.Test table
It is bright, the present invention in target scale variation, deformation, block etc. under complicated tracking scene and have good adaptability.
The above-described embodiments merely illustrate the principles and effects of the present invention, and is not intended to limit the present invention.It is any ripe
The personage for knowing this technology all without departing from the spirit and scope of the present invention, carries out modifications and changes to above-described embodiment.Cause
This, institute is complete without departing from the spirit and technical ideas disclosed in the present invention by those of ordinary skill in the art such as
At all equivalent modifications or change, should be covered by the claims of the present invention.
Claims (6)
1. a kind of dimension self-adaption nuclear phase neural network based closes filter tracking method, which is characterized in that the described method includes:
Obtain initial frame data, target area and the target area parameter information, wherein the target area parameter information
Including at least the coordinate points and size of the target area；
At least two layers of feature of the target area are extracted as characteristic pattern；
Extracted characteristic pattern is transformed into continuous domain；
By minimizing the loss function of filter response output and desired output, effective convolution filter is calculated, and be based on institute
It states effective convolution filter and carries out target prodiction；
Dimensional variation based on scaling filter prediction target；
According to the target prodiction and the target scale predicted variation, the target area of prediction is obtained, filter is found out
The affiliated example classification in the target area, joint filter are determined in the target area peak response output valve, and using classifier
Whether wave device judges target with losing in the output of the region peak response output valve and the classifier.
2. a kind of dimension self-adaption nuclear phase neural network based according to claim 1 closes filter tracking method, special
Sign is that the associated filters judge target whether with losing in the output of the region peak response output valve and the classifier
The step of losing, comprising:
Based on one multi-class classifier of MobileNetV2 network training, calculates and export the probability that target is each classification；
The maximum value is determined as target detection classification by the maximum value for determining probability；
Judge whether the peak response output valve of effective convolution filter is greater than threshold value；
If it is, indicating that target is not lost；
If not, judging whether the target detection classification is consistent with target concrete class again；
If be consistent, then it is assumed that the target is not lost, and thinks that the target is lost if not being consistent.
3. a kind of dimension self-adaption nuclear phase neural network based according to claim 1 closes filter tracking method, special
The step of sign is, the dimensional variation based on scaling filter prediction target, comprising:
Calculate the operation result of scaling filter and training sample characteristic pattern；
Calculate the mean square error with desired output；
Mean square error by minimizing desired output and calculated result acquires optimal scale filter, acquires scaling filter
Response output prediction target scale variation.
4. a kind of dimension self-adaption nuclear phase neural network based according to claim 1 closes filter tracking method, special
The step of sign is, at least two layers of feature for extracting the target area are as characteristic pattern, comprising:
Determine MobileNetV2 network inputs size；
Extract 24 conv2/1/sum layers of characteristic pattern of dimension, the 96 conv5/2/sum layers of characteristic pattern of dimension of the target area.
5. a kind of dimension self-adaption nuclear phase neural network based closes filter tracking method, feature according to claim 1
It is, described the step of extracted characteristic pattern is transformed into continuous domain, comprising:
Using interpolation operation, extracted characteristic pattern is transformed into continuous domain.
6. a kind of dimension self-adaption nuclear phase neural network based closes filter tracking method, feature according to claim 5
It is, the corresponding function of the interpolation operation embodies are as follows:
Wherein, S (γ) is interpolating function, and a is settable constant, and γ is the position of interpolation point and pixel.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811619242.3A CN109858493A (en) | 2018-12-28 | 2018-12-28 | A kind of dimension self-adaption nuclear phase pass neural network based filter tracking method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811619242.3A CN109858493A (en) | 2018-12-28 | 2018-12-28 | A kind of dimension self-adaption nuclear phase pass neural network based filter tracking method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109858493A true CN109858493A (en) | 2019-06-07 |
Family
ID=66892815
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811619242.3A Pending CN109858493A (en) | 2018-12-28 | 2018-12-28 | A kind of dimension self-adaption nuclear phase pass neural network based filter tracking method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109858493A (en) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112836684A (en) * | 2021-03-09 | 2021-05-25 | 上海高德威智能交通系统有限公司 | Auxiliary driving-based method, device and equipment for calculating degree change rate of eye scale |
| CN113327273A (en) * | 2021-06-15 | 2021-08-31 | 中国人民解放军火箭军工程大学 | Infrared target tracking method based on variable window function correlation filtering |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106887011A (en) * | 2017-01-20 | 2017-06-23 | 北京理工大学 | A kind of multi-template method for tracking target based on CNN and CF |
| US20180211396A1 (en) * | 2015-11-26 | 2018-07-26 | Sportlogiq Inc. | Systems and Methods for Object Tracking and Localization in Videos with Adaptive Image Representation |
| CN108550190A (en) * | 2018-04-19 | 2018-09-18 | 腾讯科技（深圳）有限公司 | Augmented reality data processing method, device, computer equipment and storage medium |
| CN109034193A (en) * | 2018-06-20 | 2018-12-18 | 上海理工大学 | Multiple features fusion and dimension self-adaption nuclear phase close filter tracking method |
-
2018
- 2018-12-28 CN CN201811619242.3A patent/CN109858493A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20180211396A1 (en) * | 2015-11-26 | 2018-07-26 | Sportlogiq Inc. | Systems and Methods for Object Tracking and Localization in Videos with Adaptive Image Representation |
| CN106887011A (en) * | 2017-01-20 | 2017-06-23 | 北京理工大学 | A kind of multi-template method for tracking target based on CNN and CF |
| CN108550190A (en) * | 2018-04-19 | 2018-09-18 | 腾讯科技（深圳）有限公司 | Augmented reality data processing method, device, computer equipment and storage medium |
| CN109034193A (en) * | 2018-06-20 | 2018-12-18 | 上海理工大学 | Multiple features fusion and dimension self-adaption nuclear phase close filter tracking method |
Non-Patent Citations (2)
| Title |
|---|
| MARTIN DANELLJAN ET AL.: "accurate scale estimation for robust visual tracking", 《BRITISH MACHINE VISION CONFERENCE 2014》 * |
| MARTIN DANELLJAN ET AL.: "ECO: Efficient Convolution Operators for Tracking", 《ARXIV》 * |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN112836684A (en) * | 2021-03-09 | 2021-05-25 | 上海高德威智能交通系统有限公司 | Auxiliary driving-based method, device and equipment for calculating degree change rate of eye scale |
| CN112836684B (en) * | 2021-03-09 | 2023-03-10 | 上海高德威智能交通系统有限公司 | Method, device and equipment for calculating eye scale degree change rate based on auxiliary driving |
| CN113327273A (en) * | 2021-06-15 | 2021-08-31 | 中国人民解放军火箭军工程大学 | Infrared target tracking method based on variable window function correlation filtering |
| CN113327273B (en) * | 2021-06-15 | 2023-12-19 | 中国人民解放军火箭军工程大学 | Infrared target tracking method based on variable window function correlation filtering |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110991272B (en) | Multi-target vehicle track recognition method based on video tracking | |
| CN104574445B (en) | A kind of method for tracking target | |
| CN103295242B (en) | A kind of method for tracking target of multiple features combining rarefaction representation | |
| CN110930387A (en) | Fabric defect detection method based on depth separable convolutional neural network | |
| CN107067405B (en) | Remote sensing image segmentation method based on scale optimization | |
| CN110738647B (en) | Mouse detection method integrating multi-receptive-field feature mapping and Gaussian probability model | |
| CN108305260B (en) | Method, device and equipment for detecting angular points in image | |
| CN104484890B (en) | Video target tracking method based on compound sparse model | |
| CN105787901A (en) | A multi-scale velocity field measurement method for adjacent two frames in a sun high-resolution image sequence | |
| CN111091101B (en) | High-precision pedestrian detection method, system and device based on one-step method | |
| CN104657717A (en) | Pedestrian detection method based on layered kernel sparse representation | |
| CN114627447A (en) | Road vehicle tracking method and system based on attention mechanism and multi-target tracking | |
| CN108537825B (en) | Target tracking method based on transfer learning regression network | |
| Fan et al. | SpinNet: Spinning convolutional network for lane boundary detection | |
| CN107944354A (en) | A kind of vehicle checking method based on deep learning | |
| Yang et al. | Visual tracking with long-short term based correlation filter | |
| CN109063574A (en) | A kind of prediction technique, system and the equipment of the envelope frame based on deep neural network detection | |
| CN109858493A (en) | A kind of dimension self-adaption nuclear phase pass neural network based filter tracking method | |
| CN110751670B (en) | Target tracking method based on fusion | |
| CN108257148B (en) | Target suggestion window generation method of specific object and application of target suggestion window generation method in target tracking | |
| CN113379788A (en) | Target tracking stability method based on three-element network | |
| CN112991394A (en) | KCF target tracking method based on cubic spline interpolation and Markov chain | |
| CN110111358B (en) | Target tracking method based on multilayer time sequence filtering | |
| Fan et al. | Covered vehicle detection in autonomous driving based on faster rcnn | |
| CN116597275A (en) | High-speed moving target recognition method based on data enhancement |
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 | ||
| TA01 | Transfer of patent application right | ||
| TA01 | Transfer of patent application right |
Effective date of registration: 20200601 Address after: 807c, 8th floor, Cuilin building, No. 10, Kaifeng Road, Maling community, Meilin street, Futian District, Shenzhen City, Guangdong Province Applicant after: Shenzhen Yixin Vision Technology Co.,Ltd. Address before: Room 666, No. 1199 Wuzhong Road, Minhang District, Shanghai, 201103 Applicant before: SHANGHAI READSENSE NETWORK TECHNOLOGY Co.,Ltd. |
|
| RJ01 | Rejection of invention patent application after publication | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190607 |