CN103136530A - Method for automatically recognizing target images in video images under complex industrial environment - Google Patents
Method for automatically recognizing target images in video images under complex industrial environment Download PDFInfo
- Publication number
- CN103136530A CN103136530A CN2013100419203A CN201310041920A CN103136530A CN 103136530 A CN103136530 A CN 103136530A CN 2013100419203 A CN2013100419203 A CN 2013100419203A CN 201310041920 A CN201310041920 A CN 201310041920A CN 103136530 A CN103136530 A CN 103136530A
- Authority
- CN
- China
- Prior art keywords
- images
- video image
- image
- industrial environment
- complex industrial
- 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
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for automatically recognizing target images in video images under a complex industrial environment. The method includes a first step of carrying out recognization and calculation on the video images frame by frame, a second step of converting the video images under the complex industrial environment to gray level images, a third step of carrying out luminance improvement processing or contrast ratio processing on the images through histogram equalization and histogram matching when the images are low in luminance or contrast ratio, a fourth step of carrying out de-noising processing on the images in a self-adaptive median filter mode, a fifth step of carrying out sharpening processing on the images through a Laplace filter if the sharpness is not high, a sixth step of extracting scale invariant feature transform (SIFT) feature points of the images, a sixth step of calculating matching relation between the SIFT feature points of the images and the target images, and finding key frames according with matching requirements, and a seventh step of outputting a matching result chart between the key frames and the target images. The method can effectively solves the problems that the quantity of the SIFT feature points is less and matching rate of the feature points is low caused by the low luminance and contrast ratio of the video images under the complex industrial environment, pollution of impulse noise, and image blurring.
Description
Technical field
The present invention relates to the processing technology field of video image under the complex industrial environment, under the complex industrial environment such as especially a kind of nuclear accident scene, the target images such as nuclear facilities in video image are carried out the method for identification automatically.
Background technology
Under the complex industrial environment, such as nuclear power plant will suffer huge destruction under accident conditions, compare before a lot of nuclear facilities appearances and accident and change a lot, nuclear facilities may deform or excalation, under the nuclear accident operating mode, the scene often is full of radiation in addition, is difficult to rely on manually the nuclear facilities of the scene of the accident is fast and effeciently identified.And various nuclear facilities need to be identified/locate to the alleviation of nuclear accident rapidly and accurately, thereby effectively it is assessed or operates.Therefore, adopt advanced image processing algorithm effectively to identify to the images such as target nuclear facilities in nuclear accident live video image the effective solution that becomes this difficult problem.
Under accident conditions, the nuclear facilities video image will face that the brightness and contrast is low, image is by noise pollution and the disadvantageous extraneous factor such as image blurring.Make the identification to nuclear facilities image important in the video images such as nuclear facilities become very difficult, directly affected judgement and treatment effeciency to nuclear accident.Histogram equalization conversion or coupling, adaptive median filter and Laplce's filtering operation are the comparatively Effective arithmetics that solves the unfavorable factor that nuclear accident live video image faces.These algorithms are effectively applied to greatly to improve in the identification of nuclear facilities video image accuracy and the recognition efficiency of identification.
Scale-invariant feature transform (SIFT) algorithm is a kind of widely used algorithm that carries out target identification.This algorithm is by extracting the local feature of image, and the coupling of carrying out local feature is come the identification of realize target image.Rotation, yardstick convergent-divergent, brightness are changed maintaining the invariance, visual angle change, affined transformation, noise are also kept to a certain degree stability; Be applicable to mate fast and accurately in the magnanimity property data base; But have not yet to see the histogram equalization conversion or coupling, adaptive median filter and Laplce's filtering operation combine with the SIFT algorithm and to the report of target nuclear facilities image recognition in nuclear accident live video image.
Although histogram equalization conversion or coupling, adaptive median filter and Laplce's filtering operation and SIFT algorithm have good image processing effect, but these algorithm application are when the image recognition of nuclear accident live video, need to process the view data of magnanimity, so its operand is quite large.Adopt the mode of traditional CPU and software to be difficult to satisfy the requirement of real-time of nuclear accident live video image being identified computing.
" realtime graphic based on FPGA improves algorithm of histogram equalization " that Wang Dejun etc. proposed in 2010, two histogrammic high similarity characteristics adjacent according to video image, hardware is realized improving algorithm of histogram equalization on FPGA video image processing platform, satisfies the requirement of real-time that video image is processed." based on design and the realization of the medium filtering fast algorithm of FPGA " that Wang Wei etc. 2008 propose, provided the scheme that realizes the rapid image median filtering algorithm with FPGA, good filter effect can be obtained in the image pretreatment system, and its requirement of real-time can be satisfied." the FPGA implementation method of Laplace operator " that Mao Weimin etc. proposed in 2009 designed the hardware configuration that Laplace operator realizes, has good filter effect, and design is convenient, effective.
Aspect the realization of High Speed of SIFT, HUANG F C etc. proposed " the high-performance SIFT of realtime graphic feature extraction is hardware-accelerated " scheme in 2012, realized the SIFT algorithm by two mutual mutual ASIC side hardware components.A part is used for the identification of feature key points, and another part is used for the generating feature descriptor.The Parallel Unit framework that comprises three grades of flowing water is adopted in the key point identification.When the number of unique point during less than 890, the asic chip that adopts the 180ns CMOS technique of TSMC to realize, VGA (the Video Graphic Array) image of processing a width 640*480 pixel only needs 3.4ms.And realize based on the software of 2.1GHz Intel CPU the SIFT algorithm of the VGA image of same scale needing 2.87 seconds, can't satisfy the requirement that realtime graphic is processed.Therefore, realize that based on ASIC hardware the SIFT algorithm adopts software mode to realize that speed has had huge raising.But, realize the SIFT algorithm based on asic chip, its function is fixed, and can't carry out according to the actual needs corresponding algorithm optimization, and cost high/long deficiency of construction cycle.Therefore, adopting FPGA is the optimal path of realizing the nuclear power image recognition algorithm.
In sum, at present under the complex industrial environment such as nuclear accident scene the process field of video image also do not have a kind of Effective arithmetic to carry out real-time identification to wherein image.In existing image processing algorithm, widely used SIFT algorithm is not considered the impact that the complex industrial environment such as nuclear accident scene cause video image quality, such as the contrast of image and brightness is low, image blurring or image by the situation of noise pollution.For reliably under accident conditions the video image to the nuclear accident scene carry out in real time effectively identification, thereby can the core scene after accident be operated fast and effectively, be badly in need of a kind of recognition methods with nuclear accident live video image high real-time and that can adapt to bad working environments.
Summary of the invention
Technical matters to be solved by this invention be in the complex industrial environment such as nuclear accident scene video image brightness and contrast is low, by noise pollution or the images such as identification target nuclear facilities that factor the caused difficulty such as fuzzy.
The present invention addresses the above problem the technical scheme that adopts to be: the method for automatic recognition target image in the video image under the complex industrial environment, and its step comprises:
A. the video image under the complex industrial environment is converted into gray level image frame by frame;
If b. the video image brightness of step a is low or contrast is low, adopt histogram equalization or Histogram Matching to improve brightness to it or contrast is processed;
C. the mode that adopts adaptive median filter is carried out denoising to the video image of step b;
If d. sharpness is not high, adopt Laplace filter to carry out the sharpening processing to the video image of step c;
E extracts its SIFT unique point to the video image of steps d;
F. the video image SIFT unique point of calculation procedure e acquisition and the matching relationship of target image SIFT unique point, find to meet the key frame that coupling requires;
G. export the matching result figure of key frame and target image.
Video image under described complex industrial environment is identified by above-mentioned steps a～g frame by frame.
The method of automatic recognition target image in above-mentioned video image under the complex industrial environment is characterized in that: the processing procedure of described each step all adopts FPGA to carry out.
The present invention improves the identifiability of image by the method for histogram equalization processing or Histogram Matching, adaptive median filter and Laplace transform, thereby make when identifying target nuclear facilities image in nuclear accident live video image, produce more unique point after carrying out the SIFT computing, and can improve the correct matching rate between video image and target image.Adopt FPGA to carry out each identification step and can effectively guarantee the real-time identified.
Description of drawings
Fig. 1: nuclear accident live video image is carried out the overview flow chart of identification automatically;
Fig. 2: adopt the SIFT algorithm to the process flow diagram of nuclear accident live video image automatic identification target image.
Embodiment
Below in conjunction with accompanying drawing, embodiments of the invention are described in further detail, but the present embodiment is not limited to the present invention, every employing analog structure of the present invention and similar variation thereof all should be listed protection scope of the present invention in.
The method of automatic recognition target image in video image under the complex industrial environment, its step comprises:
A. the video image under the complex industrial environment is converted into gray level image frame by frame;
If b. the video image brightness of step a is low or contrast is low, adopt histogram equalization or Histogram Matching to improve brightness to it or contrast is processed;
C. the mode that adopts adaptive median filter is carried out denoising to the video image of step b;
If d. sharpness is not high, adopt Laplace filter to carry out the sharpening processing to the video image of step c;
E. the video image of steps d extracted its SIFT unique point;
F. the video image SIFT unique point of calculation procedure e acquisition and the matching relationship of target image SIFT unique point, find to meet the key frame that coupling requires;
G. export the matching result figure of key frame and target image.
Video image under described complex industrial environment is identified by described step a～g frame by frame.
The method of automatic recognition target image in above-mentioned video image under the complex industrial environment is characterized in that: the processing procedure of described each step all adopts FPGA to carry out.
The below describes automatic identifying method of the present invention in detail take recognition target image in the video image at nuclear accident scene as example.As shown in Figure 1:
1. at step S1, with target nuclear facilities image T to be identified
bBe converted into gray level image, then carry out the SIFT computing, obtain its unique point m
bAnd feature description vectors M
b
2. at step S2, definite kernel scene of the accident video image carries out the algorithm parameter of enhance operation, and the histogram equalization that adopts or Histogram Matching, adaptive median filter and Laplce's filtering operation are the algorithm that adopts in " Digital Image Processing (Matlab version) ":
A. first three two field picture of nuclear accident live video image is converted to gray level image and carries out the SIFT computing, obtain respectively its feature n that counts
1, n
2, n
3
B. first three two field picture of nuclear accident live video image is carried out histogram equalization and process, and calculate the unique point m of its SIFT algorithm
1, m
2, m
3, make t
1=m
1-2*n
1, t
2=m
2-2*n
2, t
3=m
3-2*n
3If, t
1, t
2, t
3In two or three are arranged greater than 0, make f
1=1, otherwise make f
1=0;
C. first three two field picture of nuclear accident live video image is carried out Histogram Matching and process, and calculate the unique point p of its SIFT algorithm
1, p
2, p
3, make t
1=p
1-2*n
1, t
2=p
2-2*n
2, t
3=p
3-2*n
3If, t
1, t
2, t
3In two or three are arranged greater than 0, make f
2=1, otherwise make f
2=0;
D. first three two field picture of nuclear accident live video image is carried out Laplce's filtering, filter coefficient is [1 11; 1-4 1; 11 1], calculate the unique point r of its SIFT algorithm
1, r
2, r
3, make t
1=r
1-2*n
1, t
2=r
2-2*n
2, t
3=r
3-2*n
3If, t
1, t
2, t
3In two or three are arranged greater than 0, make f
3=1, otherwise make f
3=0.
3. at step S3, adopt the SIFT algorithm to nuclear accident live video image according to circulate the frame by frame automatic identification computing (as shown in Figure 2) of performance objective nuclear facilities image of following steps:
A. make frame number variable i=1 in nuclear accident live video image;
B. the i frame with nuclear accident live video image is converted to gray level image;
If f c.
1=1, illustrate that the i frame brightness of nuclear accident live video image is low or contrast is low, and be fit to adopt histogram equalization to process it to be processed, it is carried out histogram equalization process, generate the image T after processing
1If f
1=0; The i frame with nuclear accident live video image directly is assigned to image T
1
If f d.
2=1, illustrate that the i frame brightness of nuclear accident live video image is low or contrast is low, and be fit to adopt the Histogram Matching algorithm that it is processed, it is carried out Histogram Matching process, generate the image T after processing
2If f
2=0; With image T
1Directly be assigned to image T
2
E. to image T
2Carry out the adaptive median filter computing, generate the image T after processing
3
If f f.
3=1, illustrate that the i frame of nuclear accident live video image is fuzzy, to image T
3Carry out Laplce's filtering operation, generate the image T after processing
4If f
3=0; With image T
3Directly be assigned to image T
4
G. to image T
4Carry out the SIFT computing, obtain its unique point m
jAnd feature description vectors M
j: to image T
bFeature description vectors and image T
4The feature description vectors carry out coupling and calculate.If coupling is counted greater than 8, prove in the i frame of nuclear accident live video image to comprise target nuclear facilities image T
b, therefore export T
bAnd T
4Matching result figure.If coupling is counted less than or equal to 8, prove in the i frame of nuclear accident live video image not comprise target nuclear facilities image T
b, do not export T
bAnd T
4Matching result figure;
If h. all frames of nuclear accident live video image all are finished to the identification computing of target nuclear facilities image, stop the computing of algorithm; Otherwise make i=i+1, and go to the identification computing of next frame in b step execution nuclear accident live video image.
Claims (3)
1. the automatic method of recognition target image in the video image under the complex industrial environment, its step comprises:
A. the video image under the complex industrial environment is converted into gray level image frame by frame;
If b. the video image brightness of step a is low or contrast is low, adopt histogram equalization or Histogram Matching to improve brightness to it or contrast is processed;
C. the mode that adopts adaptive median filter is carried out denoising to the video image of step b;
If d. sharpness is not high, adopt Laplace filter to carry out the sharpening processing to the video image of step c;
E extracts its SIFT unique point to the video image of steps d;
F. the video image SIFT unique point of calculation procedure e acquisition and the matching relationship of target image SIFT unique point, find to meet the key frame that coupling requires;
G. export the matching result figure of key frame and target image.
2. the method for automatic recognition target image in the video image under the complex industrial environment according to claim 1, is characterized in that: the video image under described complex industrial environment is identified by described step a～g frame by frame.
3. the automatic method of recognition target image in the video image under the complex industrial environment according to claim 1, it is characterized in that: the processing procedure of described each step all adopts FPGA to carry out.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013100419203A CN103136530A (en) | 2013-02-04 | 2013-02-04 | Method for automatically recognizing target images in video images under complex industrial environment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN2013100419203A CN103136530A (en) | 2013-02-04 | 2013-02-04 | Method for automatically recognizing target images in video images under complex industrial environment |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103136530A true CN103136530A (en) | 2013-06-05 |
Family
ID=48496337
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN2013100419203A Pending CN103136530A (en) | 2013-02-04 | 2013-02-04 | Method for automatically recognizing target images in video images under complex industrial environment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103136530A (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106521066A (en) * | 2016-12-23 | 2017-03-22 | 天津市三特电子有限公司 | Blast furnace burden particle size monitoring system and distributed data on-line analysis method |
| CN108288041A (en) * | 2018-01-26 | 2018-07-17 | 大连民族大学 | A kind of preprocess method of pedestrian target false retrieval removal |
| CN109598706A (en) * | 2018-11-26 | 2019-04-09 | 安徽嘉拓信息科技有限公司 | A kind of camera lens occlusion detection method and system |
| WO2019223069A1 (en) * | 2018-05-25 | 2019-11-28 | 平安科技（深圳）有限公司 | Histogram-based iris image enhancement method, apparatus and device, and storage medium |
| CN110715870A (en) * | 2019-10-21 | 2020-01-21 | 张丽 | Cargo weight data cheating detection system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101819158A (en) * | 2010-03-18 | 2010-09-01 | 浙江工商大学 | Automatic detection device of flour bran speck |
| US20110013028A1 (en) * | 2009-07-17 | 2011-01-20 | Jie Zhou | Video stabilizing method and system using dual-camera system |
| CN102685460A (en) * | 2012-05-17 | 2012-09-19 | 武汉大学 | Video monitoring and cruising method for integrating measurable scene image and electronic map |
| CN102831618A (en) * | 2012-07-20 | 2012-12-19 | 西安电子科技大学 | Hough forest-based video target tracking method |
-
2013
- 2013-02-04 CN CN2013100419203A patent/CN103136530A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20110013028A1 (en) * | 2009-07-17 | 2011-01-20 | Jie Zhou | Video stabilizing method and system using dual-camera system |
| CN101819158A (en) * | 2010-03-18 | 2010-09-01 | 浙江工商大学 | Automatic detection device of flour bran speck |
| CN102685460A (en) * | 2012-05-17 | 2012-09-19 | 武汉大学 | Video monitoring and cruising method for integrating measurable scene image and electronic map |
| CN102831618A (en) * | 2012-07-20 | 2012-12-19 | 西安电子科技大学 | Hough forest-based video target tracking method |
Cited By (6)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN106521066A (en) * | 2016-12-23 | 2017-03-22 | 天津市三特电子有限公司 | Blast furnace burden particle size monitoring system and distributed data on-line analysis method |
| CN108288041A (en) * | 2018-01-26 | 2018-07-17 | 大连民族大学 | A kind of preprocess method of pedestrian target false retrieval removal |
| CN108288041B (en) * | 2018-01-26 | 2021-02-02 | 大连民族大学 | Preprocessing method for removing false detection of pedestrian target |
| WO2019223069A1 (en) * | 2018-05-25 | 2019-11-28 | 平安科技（深圳）有限公司 | Histogram-based iris image enhancement method, apparatus and device, and storage medium |
| CN109598706A (en) * | 2018-11-26 | 2019-04-09 | 安徽嘉拓信息科技有限公司 | A kind of camera lens occlusion detection method and system |
| CN110715870A (en) * | 2019-10-21 | 2020-01-21 | 张丽 | Cargo weight data cheating detection system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108230264B (en) | Single image defogging method based on ResNet neural network | |
| Xie et al. | Image edge detection based on opencv | |
| CN106845415B (en) | Pedestrian fine identification method and device based on deep learning | |
| CN111079764B (en) | Low-illumination license plate image recognition method and device based on deep learning | |
| CN110472566B (en) | High-precision fuzzy face recognition method | |
| CN103136530A (en) | Method for automatically recognizing target images in video images under complex industrial environment | |
| CN106408597A (en) | Neighborhood entropy and consistency detection-based SAR (synthetic aperture radar) image registration method | |
| CN107610093B (en) | Full-reference image quality evaluation method based on similarity feature fusion | |
| CN103400142B (en) | A kind of pedestrian counting method | |
| CN111008961A (en) | Transmission line equipment defect detection method and system, equipment and medium thereof | |
| Yan et al. | Method to Enhance Degraded Image in Dust Environment. | |
| CN103841298A (en) | Video image stabilization method based on color constant and geometry invariant features | |
| CN108197521A (en) | A kind of leggy Quick Response Code obtains identification device and method | |
| CN104537381A (en) | Blurred image identification method based on blurred invariant feature | |
| CN108711160B (en) | Target segmentation method based on HSI (high speed input/output) enhanced model | |
| CN104463091B (en) | A kind of facial image recognition method based on image LGBP feature subvectors | |
| CN106778789B (en) | A kind of fast target extracting method in multi-view image | |
| CN105654479A (en) | Multispectral image registering method and multispectral image registering device | |
| CN111027564A (en) | Low-illumination imaging license plate recognition method and device based on deep learning integration | |
| Ouzounis et al. | Interactive collection of training samples from the max-tree structure | |
| Zhang et al. | Multisensor Infrared and Visible Image Fusion via Double Joint Edge Preservation Filter and Nonglobally Saliency Gradient Operator | |
| CN104299223B (en) | Different-source image matching method based on Gabor coding | |
| CN103632357A (en) | Image super-resolution enhancing method based on illumination separation | |
| CN114582017A (en) | Generation method and generation system of gesture data set and storage medium | |
| Lanchi et al. | A novel unsharp mask sharpening method in preprocessing for face recognition |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20130605 |
|
| RJ01 | Rejection of invention patent application after publication |