CN107016353B - A kind of integrated method and system of variable resolution target detection and identification - Google Patents
A kind of integrated method and system of variable resolution target detection and identification Download PDFInfo
- Publication number
- CN107016353B CN107016353B CN201710147950.0A CN201710147950A CN107016353B CN 107016353 B CN107016353 B CN 107016353B CN 201710147950 A CN201710147950 A CN 201710147950A CN 107016353 B CN107016353 B CN 107016353B
- Authority
- CN
- China
- Prior art keywords
- target
- identification
- image
- resolution
- acquisition
- 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.)
- Active
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Abstract
A kind of integrated method and system of variable resolution target detection and identification disclosed by the invention is related to detection and identifies integrated method and system, belongs to optical image technology field.The method of the present invention includes following steps: detecting the image sensor resolutions mode with the current state setting CMOS camera for identifying integral system according to variable resolution;The image of acquisition is pre-processed, picture quality is improved;According to the current state of system, target acquisition is carried out to obtained image or target acquisition or identification are realized in Target Recognition Algorithms processing respectively;According to system current state and obtained target acquisition or target identification as a result, system mode is arranged, until realizing integrated target detection and identification.It include main control module, image interface, CMOS camera, camera lens invention additionally discloses variable resolution detection and the integrated system of identification.The technical problem to be solved by the present invention is to realize target detection and identification integration, have many advantages, such as that precision is high, small in size, strong robustness.
Description
Technical field
The invention belongs to optical image technology fields, detect more particularly to a kind of variable resolution and identify integrated side
Method and system.
Background technique
Target detection and identification technology is the non-cpntact measurement to fixed or mobile target, in the signal of measurement comprising away from
From, position, azimuth or elevation information etc., the device of this measurement can be fixation, be also possible to movement, and measure
Signal can correctly provide relevant information by special recognition methods.With highly sensitive detector, imaging sensor and
More past detection identification technology is compared in the fast development of machine vision, now accurate with farther detection range and more
Accuracy of identification.Therefore, the fields such as monitoring, navigation, Aeronautics and Astronautics be can be widely used in.
Conventional target detects the detector or sensor different from being identified by and completes, although function is more perfect,
Since different function needs to be completed by different components, lead to systems bulky, integrated level is not high.With many research fields
The integrated demand of target detection and identification is continuously increased, it is desirable that target acquisition, identification integral system have precision height, body
The advantages that product small, strong robustness.Traditional target detection and identification system by multiple sensor integrations, which has been unable to meet, to be needed
It asks.
Summary of the invention
To solve, traditional Multi-Sensor Target detection and identifying system are difficult to realize Highgrade integration and what is minimized ask
Topic, a kind of integrated method and system of variable resolution target detection and identification disclosed by the invention, technical problems to be solved
It is to realize target detection and identification integration, has many advantages, such as that precision is high, small in size, strong robustness.
The purpose of the present invention is what is be achieved through the following technical solutions:
A kind of integrated method of variable resolution target detection and identification disclosed by the invention, using variable-resolution CMOS
Camera carries out Image Acquisition, and can complete two kinds of target detection and identification methods of automated manner/manual mode.Wherein automatic side
The principle of formula are as follows: system default is in target acquisition state first, low-resolution mode is used by CMOS camera, to visual field internal field
Scape carries out Image Acquisition;Then whether judged in the visual field using target acquisition algorithm comprising suspected target;If it was found that suspected target,
Then system is automatically converted to target identification state, and uses high resolution model by CMOS camera, to scene in the visual field again at
Picture;Then using Target Recognition Algorithms determine, and according to target identification as a result, the subsequent time state of automatic adjustment system and
The resolution model of CMOS camera, to realize the automatic conversion of target detection and identification.The principle of manual mode and automatic side
Formula difference is, state (target acquisition or target identification), the resolution model of CMOS camera and the target acquisition of system with
Target identification as a result, being controlled by operator and being determined.
To realize based on the integration of variable resolution target detection and identification, adopt the following technical scheme that
A kind of integrated system of variable resolution target detection and identification disclosed by the invention, including main control module, figure
As interface, CMOS camera and camera lens.Wherein main control module by image processing algorithm realize target detection and identification, and according to
The resolution model of the result real-time control CMOS camera of target detection and identification, the resolution model are divided into for target
The low-resolution mode of detection and high resolution model for target identification.Image interface is used for CMOS camera and master control molding
Data transmission between block.CMOS camera carries out Image Acquisition by photoelectric conversion, to current goal, and CMOS chip need to have
Resolution ratio adjustable function.Camera lens is used to control the field angle and image-forming range of CMOS camera.
A kind of integrated method of variable resolution target detection and identification disclosed by the invention, includes the following steps:
Step 1: it is detected according to variable resolution and is passed with the image for the current state setting CMOS camera for identifying integral system
Sensor resolution model.
It is detected according to variable resolution and identifies integrated system (hereinafter referred to as system) current state, CMOS camera is set
Image sensor resolutions mode.If system is in target acquisition state, it is set as low-resolution mode, the low resolution
Rate mode is realized by merging the method for adjacent pixel;If system is in target identification state, then high resolution model is set as.
The described variable resolution detection with identify integrated system include main control module, image interface, CMOS camera,
Camera lens.Wherein main control module realizes target detection and identification by image processing algorithm, and according to target detection and identification
As a result the resolution model of real-time control CMOS camera, the resolution model are divided into the low resolution mould for target acquisition
Formula and high resolution model for target identification.Image interface is passed for the data between CMOS camera and main control module
It is defeated.CMOS camera carries out Image Acquisition by photoelectric conversion, to current goal, and CMOS chip need to have resolution ratio can Power Regulation
Energy.Camera lens is used to control the field angle and image-forming range of CMOS camera.
Step 2: pre-processing the image of acquisition, to remove noise, improves picture quality.
Step 3: according to the current state of system, target acquisition is carried out to the image that step 2 obtains or target identification is calculated
Target acquisition or identification are realized in method processing respectively.
If the current state of system is target acquisition, operational objective probe algorithm realizes target acquisition.Target acquisition is calculated
Method uses the vision significance model (Graph-Based Visual Saliency, GBVS) based on graph theory to realize image first
Significance analysis, obtain notable figure, and carry out salient region segmentation using region-growing method, obtain area-of-interest;So
Afterwards using the grey level histogram of area-of-interest and target template image as characteristics of image, normalizated correlation coefficient, meter are calculated
Calculate formula are as follows:
Wherein: α and β is respectively the feature vector of area-of-interest and template,WithThe respectively mean value of vector α and β.
It is robustness of the raising system to illumination variation, first with correlation since grey level histogram feature is sensitive to illumination variation
Matching algorithm is registrated two vectors, obtains the best match position of two vectors, and carry out vector pair according to the position
Standard, then calculate the normalizated correlation coefficient γ (α, β) of two feature vectors.Resulting normalizated correlation coefficient γ (α, β) is used for
Suspected target is judged whether there is, i.e. completion target acquisition.
If the current state of system is target identification, operational objective recognizer realizes target identification.Target identification is calculated
Method carries out region of interest regional partition to current frame image first, obtains new area-of-interest;Then sparse coding algorithm is utilized,
Sparse coding is carried out to area-of-interest, gained vector is the feature vector of respective image;Then feature vector is input to
Trained support vector machines under line, acquired results find real goal, i.e. completion target identification for judging whether.
Step 4: the target acquisition or target identification obtained according to system current state and step 3 is as a result, setting system
State, until realizing integrated target detection and identification based on variable resolution.
When system is in target acquisition state, if the target acquisition algorithm of step 3 finds suspected target, system
Into target identification state, return step one;If not finding suspected target, system continues to keep target acquisition state, returns
Step 1.When system is in target identification state, if not finding real goal by step 3, system enters target acquisition
State, return step one;If it find that real goal, then system enters target following task or other tasks, that is, is based on variation
Resolution realizes integrated target detection and identification.
The result of target acquisition described in step 3 or target identification is by either automatically or manually judging whether to deposit
In suspected target or real goal.
Suspected target or real goal concrete methods of realizing are judged whether there is by automated manner are as follows: target is visited
It surveys, obtains normalizated correlation coefficient by the target acquisition algorithm, and it is compared with preset threshold value, acquired results conduct
Unique classification standard judges automatically and whether finds suspected target;For target identification, first with the interested of previous frame image
The area-of-interest that region contour extracts present frame obtains the output of support vector machines then according to the Target Recognition Algorithms
As a result, and as unique classification standard, judge automatically and whether find real goal, i.e., realization automated manner judge whether to deposit
In suspected target or real goal.
Suspected target or real goal are judged whether there is by automated manner, since system enters target identification shape every time
State only handles single-frame images and is just converted to other states, therefore simplifies the region of interesting extraction algorithm of target identification, improves
System effectiveness, the resolution ratio and working condition of this external system can realize automatic switchover, improve the intelligence of system.
Manually judge whether there is suspected target or real goal concrete methods of realizing are as follows: target is visited
It surveys, obtains normalizated correlation coefficient by the target acquisition algorithm, and using the coefficient as subsidiary classification standard shows to image
On, finally decided whether to find suspected target by operator;For target identification, first with vision significance as described above
Model and region-growing method extract area-of-interest, then obtain feature vector with sparse coding algorithm, are then input under line and instruct
The support vector machines perfected simultaneously obtains its calculated result, and using the result as in subsidiary classification standard shows to image, finally
Decided whether to find real goal by operator, i.e. realization manual mode judges whether there is suspected target or real goal.
Suspected target or real goal are manually judged whether there is, this method not only makes full use of training under line
Priori knowledge, and combine the distinctive priori knowledge of operator, reduce systematic error, the priori for compensating for training under line is known
Know insufficient defect, improves the robustness and precision of system.
The concrete methods of realizing of step three Target Recognition Algorithms includes the following steps:
Since using support vector machines, as target identification classifier, the specific implementation of target identification is by training under line
It is formed with two parts are tested on line.The specific method is as follows for the process of training under target identification line:
Step (1): all training set images containing label are handled using significance analysis algorithm same as above.
Step (2): using region-growing method same as above, carries out salient region extraction to training set image, obtains
Training set area-of-interest.
Step (3): utilizing sparse coding algorithm, carries out sparse coding to obtained each training set area-of-interest,
Gained vector is the feature vector of respective image.
Step (4): using the resulting feature vector of step (3) and its corresponding label as input, Training Support Vector Machines,
And using trained support vector machines as the target identification classifier tested on line.
The specific method is as follows for testing process on the line of target identification:
Step (1): salient region extraction is carried out to image obtained by step 2, obtains area-of-interest.
Step (2): area-of-interest obtained in step (1) is felt using sparse coding algorithm as described above
The feature vector in interest region.
Step (3): inputting trained support vector machines for feature vector obtained in step (2), obtains output result.
Step (4): based on step (3) resulting output as a result, using either automatically or manually judging whether there is
Suspected target or real goal.
The utility model has the advantages that
1, a kind of variable resolution detection disclosed by the invention and the integrated method and system of identification, can be according to different need
Ask, by determine resolution image sensor carry out pixel combination meet detection with identification demand, have precision it is high, it is small in size,
The advantages that strong robustness.The method that automated manner and manual mode combine carries out the judgement of target presence or absence, can be improved
The efficiency of system, and can make up for it the incomplete defect of simple automated manner priori knowledge, improve target acquisition and identification essence
Degree;Detection is realized using variable-resolution image sensor and identifies integrated method, simplifies target acquisition and identifying system
Structure, be conducive to the miniaturization of system;The designed target acquisition algorithm based on grey level histogram causes illumination variation
Changing features have robustness.
2, a kind of variable resolution detection disclosed by the invention and the integrated method and system of identification, with manually and automatically
Mode, manual mode can be docked with the imaging with priori knowledge, automatic mode can with do not have priori knowledge or pass through line
The image data base of lower training is docked, and therefore, present system scalability is good, easily scalable.
3, a kind of variable resolution detection disclosed by the invention and the integrated method and system of identification, can be adaptively from target
Identification transition is detected, also only work there can be stronger versatility in detection or recognition mode.
Detailed description of the invention
Fig. 1 is a kind of variable resolution detection disclosed by the invention and the integrated system construction drawing of identification;
Fig. 2 is a kind of work flow diagram variable resolution detection and identify integrated system disclosed by the invention;
Fig. 3 is detection submodule flow chart;
Fig. 4 is identification submodule flow chart;
Fig. 5 is that former resolution ratio samples schematic diagram;
Fig. 6 is that variable resolution samples schematic diagram (2 × 2);
Fig. 7 is that variable resolution samples schematic diagram (4 × 4).
Wherein: 1- main control module, 2- image interface, 3-CMOS camera, 4- camera lens.
Specific embodiment
Below in conjunction with attached drawing, description of specific embodiments of the present invention.
Embodiment 1:
The present embodiment realizes a kind of integrated method and system of variable resolution target detection and identification using automated manner.
As shown in Figure 1, a kind of integrated system of variable resolution target detection and identification disclosed in the present embodiment includes master
Control module 1, image interface 2, CMOS camera 3 and camera lens 4.Wherein main control module 1 is for realizing target detection and identification,
And the resolution model of real-time control CMOS camera 3.The imaging sensor of CMOS camera 3 should have the function of pixel combination, this
The imaging sensor model MT9V032 that embodiment uses.
As shown in Fig. 2, the present embodiment realizes that a kind of variable resolution target detection and identification is integrated using automated manner
Method includes the following steps:
Step 1: it is detected according to variable resolution and the image of CMOS camera 3 is set with integrated system current state is identified
Sensor resolution mode.
It is detected according to variable resolution and identifies integrated system (hereinafter referred to as system) current state, CMOS camera is set
3 image sensor resolutions mode.If system is in target acquisition state, it is set as low-resolution mode, described low point
Resolution mode is realized by merging the method for adjacent pixel;If system is in target identification state, then high resolution model is set as.
The variable resolution detection and the integrated system of identification include main control module 1, image interface 2, CMOS phase
Machine 3, camera lens 4.Wherein main control module 1 by image processing algorithm realize target detection and identification, and according to target acquisition with
The resolution model of the result real-time control CMOS camera 3 of identification, the resolution model are divided into for the low of target acquisition
Resolution model and high resolution model for target identification.Image interface is used between CMOS camera 3 and main control module 1
Data transmission.CMOS camera 3 carries out Image Acquisition by photoelectric conversion, to current goal, and CMOS chip need to have resolution
Rate adjustable function.Camera lens is used to control the field angle and image-forming range of CMOS camera 3.
Step 2: pre-processing the image of acquisition, to remove noise, improves picture quality.
Step 3: according to the current state of system, target acquisition is carried out to the image that step 2 obtains or target identification is calculated
Target acquisition or identification are realized in method processing respectively.
If the current state of system is target acquisition, operational objective probe algorithm realizes target acquisition.Target acquisition is calculated
Method uses the significance analysis of vision significance model realization image as described above first, obtains notable figure, and utilize region
Growth method carries out salient region segmentation, obtains area-of-interest;Then with the gray scale of area-of-interest and target template image
Histogram calculates normalizated correlation coefficient as characteristics of image, its calculation formula is:
Wherein: α and β is respectively the feature vector of area-of-interest and template,WithThe respectively mean value of vector α and β.
It is robustness of the raising system to illumination variation, first with correlation since grey level histogram feature is sensitive to illumination variation
Matching algorithm is registrated two vectors, obtains the best match position of two vectors, and carry out vector pair according to the position
Standard, then calculate the normalizated correlation coefficient γ (α, β) of two feature vectors.Resulting normalizated correlation coefficient γ (α, β) is used for
Suspected target is judged whether there is, i.e. completion target acquisition.
If the current state of system is target identification, operational objective recognizer realizes target identification.Target identification is calculated
The area-of-interest profile of previous frame image is first mapped to present image by method, to realize that salient region is divided, obtains feeling emerging
Interesting region;Then sparse coding algorithm is utilized, sparse coding is carried out to area-of-interest, gained vector is the spy of respective image
Levy vector;Then feature vector is input to trained support vector machines under line, is judged automatically and whether is sent out using acquired results
Existing real goal, i.e. completion target identification.
Step 4: the target acquisition or target identification obtained according to system current state and step 3 is as a result, setting system
State.
When system is in target acquisition state, if the target acquisition algorithm of step 3 finds suspected target, system
Into target identification state, return step one;If not finding suspected target, system continues to keep target acquisition state, returns
Step 1.When system is in target identification state, if not finding real goal by step 3, system enters target acquisition
State, return step one;If it find that real goal, then system enters target following task or other tasks, that is, is based on variation
Resolution realizes integrated target detection and identification.
Embodiment 2:
The present invention realizes a kind of integrated method and system of variable resolution target detection and identification using manual mode.
As shown in Figure 1, a kind of integrated system of variable resolution target detection and identification disclosed in the present embodiment includes master
Control module 1, image interface 2, CMOS camera 3 and camera lens 4.Wherein main control module 1 is for realizing target detection and identification,
And the resolution model of real-time control CMOS camera 3.The imaging sensor of CMOS camera 3 should have the function of pixel combination, this
The imaging sensor model MT9V032 that embodiment uses.
As shown in Fig. 2, the present embodiment realizes that a kind of variable resolution target detection and identification is integrated using manual mode
Method includes the following steps:
Step 1: it is detected according to variable resolution and the image of CMOS camera 3 is set with integrated system current state is identified
Sensor resolution mode.
It is detected according to variable resolution and identifies integrated system (hereinafter referred to as system) current state, CMOS camera is set
3 image sensor resolutions mode.If system is in target acquisition state, it is set as low-resolution mode, described low point
Resolution mode is realized by merging the method for adjacent pixel;If system is in target identification state, then high resolution model is set as.
The variable resolution detection and the integrated system of identification include main control module 1, image interface 2, CMOS phase
Machine 3, camera lens 4.Wherein main control module 1 by image processing algorithm realize target detection and identification, and according to target acquisition with
The resolution model of the result real-time control CMOS camera 3 of identification, the resolution model are divided into for the low of target acquisition
Resolution model and high resolution model for target identification.Image interface is used between CMOS camera 3 and main control module 1
Data transmission.CMOS camera 3 carries out Image Acquisition by photoelectric conversion, to current goal, and CMOS chip need to have resolution
Rate adjustable function.Camera lens is used to control the field angle and image-forming range of CMOS camera 3.
Step 2: pre-processing the image of acquisition, to remove noise, improves picture quality.
Step 3: according to the current state of system, target acquisition is carried out to the image that step 2 obtains or target identification is calculated
Target acquisition or identification are realized in method processing respectively.
If the current state of system is target acquisition, operational objective probe algorithm realizes target acquisition.Target acquisition is calculated
Method uses the significance analysis of vision significance model realization image as described above first, obtains notable figure, and utilize region
Growth method carries out salient region segmentation, obtains area-of-interest;Then with the gray scale of area-of-interest and target template image
Histogram calculates normalizated correlation coefficient as characteristics of image, its calculation formula is:
Wherein: α and β is respectively the feature vector of area-of-interest and template,WithThe respectively mean value of vector α and β.
It is robustness of the raising system to illumination variation, first with correlation since grey level histogram feature is sensitive to illumination variation
Matching algorithm is registrated two vectors, obtains the best match position of two vectors, and carry out vector pair according to the position
Standard, then calculate the normalizated correlation coefficient γ (α, β) of two feature vectors.By resulting normalizated correlation coefficient γ (α, β) with
Preset threshold value compares, and comparing result is shown on image, determines whether to find suspected target by operator, i.e., complete
At target acquisition.
If the current state of system is target identification, operational objective recognizer realizes target identification.Target identification is calculated
Method carries out salient region segmentation to image first with vision significance model as described above and region-growing method, obtains feeling emerging
Interesting region;Then sparse coding algorithm is utilized, sparse coding is carried out to area-of-interest, gained vector is the spy of respective image
Levy vector;Then feature vector is input to trained support vector machines under line, when acquired results are real goal, then existed
The area-of-interest is marked with green outline in image, is otherwise marked with red contours.For whether finding real goal by grasping
Author determines, that is, completes target identification.
Step 4: the target acquisition or target identification obtained according to system current state and step 3 is as a result, setting system
State.
When system is in target acquisition state, if in step 3 operator determine discovery suspected target, system into
Enter target identification state, return step one;If operator determines not finding that suspected target or operator do not determine,
System continues to keep target acquisition state, return step one.When system is in target identification state, if operator does not make
Determine, then system continues to keep target identification state, return step one;If operator determines not finding true mesh in step 3
Mark, then system enters target acquisition state, return step one;If operator determines discovery real goal, system enters mesh
Tracing task or other tasks are marked, i.e. integrated target detection and identification of the realization based on variable resolution.
The above is only preferred embodiments of the invention, are not intended to limit the scope of the present invention.It is all in the present invention
Spirit and principle within, any modification, equivalent replacement, improvement and so on, should be included in protection scope of the present invention it
It is interior.
Claims (5)
1. a kind of integrated method of variable resolution target detection and identification, characterized by the following steps:
Step 1: it is detected according to variable resolution and is passed with the image for current state setting CMOS camera (3) for identifying integral system
Sensor resolution model;
It is detected according to variable resolution and identifies integrated system (hereinafter referred to as system) current state, be arranged CMOS camera (3)
Image sensor resolutions mode;If system is in target acquisition state, it is set as low-resolution mode, the low resolution
Rate mode is realized by merging the method for adjacent pixel;If system is in target identification state, then high resolution model is set as;
Step 2: pre-processing the image of acquisition, to remove noise, improves picture quality;
Step 3: according to the current state of system, the image that step 2 obtains is carried out at target acquisition or Target Recognition Algorithms
Reason realizes target acquisition or identification respectively;
If the current state of system is target acquisition, operational objective probe algorithm realizes target acquisition;Target acquisition algorithm is first
The aobvious of image is first realized using the vision significance model (Graph-Based Visual Saliency, GBVS) based on graph theory
The analysis of work property obtains notable figure, and carries out salient region segmentation using region-growing method, obtains area-of-interest;Then with
The grey level histogram of area-of-interest and target template image calculates normalizated correlation coefficient as characteristics of image, calculates public
Formula are as follows:
Wherein: α and β is respectively the feature vector of area-of-interest and template,WithThe respectively mean value of vector α and β;Due to
Grey level histogram feature is sensitive to illumination variation, therefore is raising system to the robustness of illumination variation, first with relevant matches
Algorithm is registrated two vectors, obtains the best match position of two vectors, and carry out vector alignment according to the position, then
Calculate the normalizated correlation coefficient γ (α, β) of two feature vectors;Resulting normalizated correlation coefficient γ (α, β) is used to judge
It is no that there are suspected target, i.e. completion target acquisitions;
If the current state of system is target identification, operational objective recognizer realizes target identification;Target Recognition Algorithms are first
The salient region segmentation for first extracting input picture, obtains new area-of-interest;Then sparse coding algorithm is utilized, it is emerging to feeling
Interesting region carries out sparse coding, and gained vector is the feature vector of respective image;Then feature vector is input under line and is instructed
The support vector machines perfected, acquired results find real goal, i.e. completion target identification for judging whether;
Step 4: the target acquisition or target identification that are obtained according to system current state and step 3 as a result, setting system mode,
Until realizing integrated target detection and identification based on variable resolution;
When system is in target acquisition state, if the target acquisition algorithm of step 3 finds suspected target, system enters
Target identification state, return step one;If not finding suspected target, system continues to keep target acquisition state, return step
One;When system is in target identification state, if not finding real goal by step 3, system enters target acquisition shape
State, return step one;If it find that real goal, then system enters target following task, i.e., realizes one based on variable resolution
Change target detection and identification.
2. a kind of integrated method of variable resolution target detection and identification according to claim 1, it is characterised in that:
The detection of variable resolution described in step 1 with identify integrated system include main control module (1), image interface (2),
CMOS camera (3), camera lens (4);Wherein main control module (1) realizes target detection and identification, and root by image processing algorithm
According to the resolution model of the result real-time control CMOS camera (3) of target detection and identification, the resolution model is divided into use
Low-resolution mode in target acquisition and the high resolution model for target identification;Image interface (2) is used for CMOS camera
(3) the data transmission between main control module (1);CMOS camera (3) is carried out image to current goal and is adopted by photoelectric conversion
Collection, and CMOS chip need to have resolution ratio adjustable function;Camera lens is used to control the field angle and image-forming range of CMOS camera (2).
3. a kind of integrated method of variable resolution target detection and identification according to claim 1 or 2, feature exist
In:
The result of target acquisition described in step 3 or target identification is doubtful by either automatically or manually judging whether there is
Like target or real goal;
Suspected target or real goal concrete methods of realizing are judged whether there is by automated manner are as follows: for target acquisition, are pressed
The target acquisition algorithm obtains normalizated correlation coefficient, and it is compared with preset threshold value, and acquired results are as unique
Classification standard judges automatically and whether finds suspected target;For target identification, first with the area-of-interest of previous frame image
The area-of-interest of contours extract present frame, then according to the target acquisition algorithm, obtain the output of support vector machines as a result,
And as unique classification standard, judge automatically and whether find real goal, i.e., realization automated manner judges whether there is doubtful
Like target or real goal;
Manually judge whether there is suspected target or real goal concrete methods of realizing are as follows: for target acquisition, press
The target acquisition algorithm obtains normalizated correlation coefficient, and using the coefficient as in subsidiary classification standard shows to image, most
Decided whether to find suspected target by operator eventually;For target identification, first with vision significance model as described above
Area-of-interest is extracted with region-growing method, then obtains feature vector with sparse coding algorithm, is then input under line and trains
Support vector machines and obtain its calculated result, and using the result as in subsidiary classification standard shows to image, finally by grasping
Author decides whether discovery real goal, i.e. realization manual mode judges whether there is suspected target or real goal.
4. a kind of integrated method of variable resolution target detection and identification according to claim 1 or 2, feature exist
In:
Since using support vector machines, as target identification classifier, the specific implementation of target identification is by trained and line under line
Upper test two parts composition;The specific method is as follows for the process of training under target identification line:
Step (1): all training set images containing label are handled using significance analysis algorithm same as above;
Step (2): using region-growing method same as above, carries out salient region extraction to training set image, is trained
Collect area-of-interest;
Step (3): utilizing sparse coding algorithm, carries out sparse coding, gained to obtained each training set area-of-interest
Vector is the feature vector of respective image;
Step (4): it using the resulting feature vector of step (3) and its corresponding label as input, Training Support Vector Machines, and incites somebody to action
Trained support vector machines is as the target identification classifier tested on line;
The specific method is as follows for testing process on the line of target identification:
Step (1): salient region extraction is carried out to image obtained by step 2, obtains area-of-interest;
Step (2): area-of-interest obtained in step (1) is obtained interested using sparse coding algorithm as described above
The feature vector in region;
Step (3): inputting trained support vector machines for feature vector obtained in step (2), obtains output result;
Step (4): based on step (3) resulting output as a result, doubtful using either automatically or manually judging whether there is
Target or real goal.
5. the system for realizing the method as described in Claims 1-4 any one, it is characterised in that: including main control module
(1), image interface (2), CMOS camera (3), camera lens (4);Wherein main control module (1) realizes target by image processing algorithm
Detection and identification, and the resolution model of the result real-time control CMOS camera (3) according to target detection and identification, point
Resolution mode is divided into the low-resolution mode for target acquisition and the high resolution model for target identification;Image interface
(2) for the data transmission between CMOS camera (3) and main control module (1);CMOS camera (3) is by photoelectric conversion, to working as
Preceding target carries out Image Acquisition, and CMOS chip need to have resolution ratio adjustable function;Camera lens is used to control the view of CMOS camera (2)
Rink corner and image-forming range.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710147950.0A CN107016353B (en) | 2017-03-13 | 2017-03-13 | A kind of integrated method and system of variable resolution target detection and identification |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201710147950.0A CN107016353B (en) | 2017-03-13 | 2017-03-13 | A kind of integrated method and system of variable resolution target detection and identification |
Publications (2)
Publication Number | Publication Date |
---|---|
CN107016353A CN107016353A (en) | 2017-08-04 |
CN107016353B true CN107016353B (en) | 2019-08-23 |
Family
ID=59440861
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201710147950.0A Active CN107016353B (en) | 2017-03-13 | 2017-03-13 | A kind of integrated method and system of variable resolution target detection and identification |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN107016353B (en) |
Families Citing this family (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN107505628B (en) * | 2017-08-15 | 2020-06-16 | 北京理工大学 | Optical phased array variable resolution imaging system and method |
WO2020034083A1 (en) * | 2018-08-14 | 2020-02-20 | Huawei Technologies Co., Ltd. | Image processing apparatus and method for feature extraction |
CN110458198B (en) * | 2019-07-10 | 2022-03-29 | 哈尔滨工业大学(深圳) | Multi-resolution target identification method and device |
CN111553262B (en) * | 2020-04-26 | 2023-09-01 | 上海微阱电子科技有限公司 | Detection device and method for rapidly detecting target graph |
CN111798634B (en) * | 2020-06-29 | 2022-02-01 | 杭州海康威视数字技术股份有限公司 | Perimeter detection method and device |
CN112987291A (en) * | 2021-04-09 | 2021-06-18 | 西安文理学院 | Novel variable-resolution optical system for eliminating aberration by using deformable mirror and imaging method thereof |
Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104819991A (en) * | 2015-05-15 | 2015-08-05 | 北京华力兴科技发展有限责任公司 | Radiation imaging method, system and device capable of changing resolution ratio |
CN204679638U (en) * | 2015-06-24 | 2015-09-30 | 武汉万集信息技术有限公司 | A kind of laser range sensor of variable sweep resolution |
CN105158769A (en) * | 2015-07-29 | 2015-12-16 | 北京理工大学 | Double-linkage bionic eye laser scanning imaging system based on MOEMS device |
CN105898308A (en) * | 2015-12-18 | 2016-08-24 | 乐视云计算有限公司 | Resolution-variable coding mode prediction method and device |
-
2017
- 2017-03-13 CN CN201710147950.0A patent/CN107016353B/en active Active
Patent Citations (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN104819991A (en) * | 2015-05-15 | 2015-08-05 | 北京华力兴科技发展有限责任公司 | Radiation imaging method, system and device capable of changing resolution ratio |
CN204679638U (en) * | 2015-06-24 | 2015-09-30 | 武汉万集信息技术有限公司 | A kind of laser range sensor of variable sweep resolution |
CN105158769A (en) * | 2015-07-29 | 2015-12-16 | 北京理工大学 | Double-linkage bionic eye laser scanning imaging system based on MOEMS device |
CN105898308A (en) * | 2015-12-18 | 2016-08-24 | 乐视云计算有限公司 | Resolution-variable coding mode prediction method and device |
Non-Patent Citations (1)
Title |
---|
一种空间变分辨率传感系统设计与实现;刘越;《中国优秀硕士学位论文库》;20150715;第10,19,25页 |
Also Published As
Publication number | Publication date |
---|---|
CN107016353A (en) | 2017-08-04 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN107016353B (en) | A kind of integrated method and system of variable resolution target detection and identification | |
CN107229930B (en) | Intelligent identification method for numerical value of pointer instrument | |
CN106056597B (en) | Object visible detection method and device | |
JP6305171B2 (en) | How to detect objects in a scene | |
CN106886216B (en) | Robot automatic tracking method and system based on RGBD face detection | |
CN106682603B (en) | Real-time driver fatigue early warning system based on multi-source information fusion | |
US20110025834A1 (en) | Method and apparatus of identifying human body posture | |
CN106529559A (en) | Pointer-type circular multi-dashboard real-time reading identification method | |
CN102737370B (en) | Method and device for detecting image foreground | |
KR20140045854A (en) | Method and apparatus for monitoring video for estimating gradient of single object | |
CN110096980A (en) | Character machining identifying system | |
CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
CN109359577A (en) | A kind of Complex Background number detection system based on machine learning | |
CN109685038A (en) | A kind of article clean level monitoring method and its device | |
CN101320477B (en) | Human body tracing method and equipment thereof | |
CN108073940A (en) | A kind of method of 3D object instance object detections in unstructured moving grids | |
CN102201060B (en) | Method for tracking and evaluating nonparametric outline based on shape semanteme | |
CN116703895B (en) | Small sample 3D visual detection method and system based on generation countermeasure network | |
CN109443319A (en) | Barrier range-measurement system and its distance measuring method based on monocular vision | |
CN110334727B (en) | Intelligent matching detection method for tunnel cracks | |
CN110516527B (en) | Visual SLAM loop detection improvement method based on instance segmentation | |
CN114662594B (en) | Target feature recognition analysis system | |
CN110322508A (en) | A kind of assisted location method based on computer vision | |
CN109472223A (en) | A kind of face identification method and device | |
CN106355132B (en) | Face static state skin area automatic identification detection method and its system |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |