CN105352482B - 332 dimension object detection methods and system based on bionic compound eyes micro lens technology - Google Patents
332 dimension object detection methods and system based on bionic compound eyes micro lens technology Download PDFInfo
- Publication number
- CN105352482B CN105352482B CN201510732346.5A CN201510732346A CN105352482B CN 105352482 B CN105352482 B CN 105352482B CN 201510732346 A CN201510732346 A CN 201510732346A CN 105352482 B CN105352482 B CN 105352482B
- Authority
- CN
- China
- Prior art keywords
- mrow
- image
- msub
- target
- object detection
- 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
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C11/00—Photogrammetry or videogrammetry, e.g. stereogrammetry; Photographic surveying
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01B—MEASURING LENGTH, THICKNESS OR SIMILAR LINEAR DIMENSIONS; MEASURING ANGLES; MEASURING AREAS; MEASURING IRREGULARITIES OF SURFACES OR CONTOURS
- G01B11/00—Measuring arrangements characterised by the use of optical techniques
- G01B11/24—Measuring arrangements characterised by the use of optical techniques for measuring contours or curvatures
Landscapes
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- Image Processing (AREA)
Abstract
The present invention relates to a kind of 332 dimension object detection methods and system based on bionic compound eyes micro lens technology, seizure imaging is carried out to target area using the high-resolution data obtaining mode based on bionic compound eyes structure microlens system, according to the microlens array image that two lenticule devices are shot using linear weighted function method of average structure low resolution image；Using the three-D profile of forward intersection measuring method reconstruct target；If after effectively capturing target in low resolution image, the data based on microlens array image, the high resolution image of target area is reconstructed using the method for regularization；After the high resolution 2 d image for obtaining target area, target is accurately identified using the GAC models based on texture gradient.The step of present invention captures target three-D profile by increasing by one using low resolution image, effectively avoids the real-time treatment effeciency and accuracy for the insignificant processing of redundancy image, improving system.
Description
Technical field
It is particularly a kind of micro- based on bionic compound eyes the present invention relates to a kind of method and system for improving target detection efficiency
The 3-3-2 dimension object detection methods and system of mirror technology, that is, utilize the two-level resolution data acquisition pair of microlens system
Target carries out high accuracy, the high efficiency object detection method of the capture of low resolution three-D profile and high resolution 2 d staring imaging
And system.
Background technology
Traditional object detection method is that target is shot based on single imaging device, in the magnanimity high-resolution of acquisition
Target detection is carried out on rate image.The image of camera shooting is more, the higher data for meaning to obtain of the imaging resolution of camera
Amount is bigger, data message is more.This cause computer carry out automatic business processing time it is longer, efficiency is lower.In addition, target position
The uncertain surely effective capture movement target that causes to differ in high-definition picture of confidence breath, i.e., in the presence of substantial amounts of redundancy shadow
Picture.Traditional method, which does not take into account that, whether there is target in image, but unitized processing is carried out to image.This certainly will cause to count
The efficiency that calculation machine carries out target detection reduces.Current researcher concentrates on goal in research detection algorithm.Although scientific research work
The algorithm that author proposes can improve the target detection efficiency of algorithm, but the redundant data of magnanimity result in the real-time processing of system
Efficiency is relatively low.How to avoid carrying out insignificant processing to redundancy image, research both domestic and external is also nearly at space state.
The content of the invention
In order to overcome conventional method to carry out insignificant processing to the redundancy image of magnanimity, the invention provides a kind of base
Object detection method and system are tieed up in the 3-3-2 of bionic compound eyes micro lens technology.
To achieve the above object, the present invention uses following technical scheme：
Two-level resolution data acquisition is built using microlens system, carrying out low resolution three-D profile to target catches
Obtain with high resolution 2 d staring imaging, the step of capturing target three-D profile using low resolution image by increasing by one,
Target can more efficiently be judged whether in the target area, effectively avoid to the insignificant processing of redundancy image, improve
The real-time treatment effeciency and accuracy of system.
Specifically, the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, comprise the following steps：
1) seizure imaging is carried out to target area using the microlens system based on bionic compound eyes structure as imaging system, will
Obtained microlens array image is shot using linear weighted function method of average reconstruct low resolution image；
2) in step 1) based on the low resolution image of reconstruct, target point is calculated using forward intersection measuring method
Three-dimensional coordinate, low resolution three-D profile capture is carried out to target；
If 3) effectively after capture target, data based on the microlens array image obtained by step 1) shooting, use
Regularization method reconstructs high resolution image, and high resolution 2 d staring imaging is carried out to target area；Otherwise, mobile lenticule
System simultaneously returns to step 1)；
4) after the high resolution 2 d image for obtaining target area, target is entered using the GAC models based on texture gradient
The accurate identification of row, completes target detection.
In step 1), the described linear weighted function method of average is as follows：
Wherein,
In formula, the pixel in unit image is ranked up from small to large according to gray value, and it is 1 to m to be numbered, gm
Represent the gray value of the pixel that numbering is m in unit image, piFor gray value giCorresponding weight, niRepresent that gray value is
giNumber of pixels.
Step 2) specifically includes herein below：1. feature point extraction is carried out to image with matching；2. screened in match point
Part same place carries out relative orientation, that is, determines the relative attitude information between lenticule device；3. according between lenticule device
Relative attitude information, using forward intersection measuring method obtain match point relative dimensional coordinate, that is, obtain target three-dimensional
Profile.
The algorithm determines the principal direction of each characteristic point using topography's Gradient Features of characteristic point, and its formula is such as
Under：
(x, y) is characterized coordinate a little in formula, and m (x, y) and θ (x, y) are respectively the gaussian pyramid image of current scale
Gradient and direction at (x, y) place, L (x, y) are gray scale of the gaussian pyramid image at (x, y) place of current scale.
The relative attitude information between two lenticule devices is determined, that is, calculates the elements of relative orientation of two photosEquation be:
Wherein,Wherein Q is to regard up and down
Difference, N1, N2For projection coefficient, (X1,Y1,Z1), (X2,Y2,Z2) it is coordinate of the picture point in the auxiliary coordinates of image space, BX,BY,BZ
It is projection of the photographic base on XYZ directions, d is differential sign.Using the principle of least square, can establish an equation
In formula, l be indirect adjustment free term, an,bn,cn,dn,enFor the coefficient of error equation, vnFor indirect adjustment
Error term, elements of relative orientation can be tried to achieve by indirect adjustmentWhereinIt is relative for second photo
The elements of interior orientation of first photo, (μ, ν) are the drift angle and inclination angle of baseline.
Comprising the following steps that for the relative dimensional coordinate of match point is obtained using forward intersection measuring method：Calculate first
Angular range element and baseline component (BX、BY、BZ)；Calculate orthogonal matrix of the left and right photo in photogrammetric coordinate system；Calculate picture
Coordinate (X of the point in the auxiliary coordinates of image space1、Y1、Z1) and (X2、Y2、Z2)；Calculate projection coefficient N1、N2；Computation model point
Three-dimensional coordinate (X, Y, Z).Three-dimensional coordinate calculation formula of the model points in the auxiliary coordinates of image space is as follows：
The three-dimensional coordinate that substantial amounts of model points are obtained by the above method obtains the three-D profile of target.
In step 3), the method that high resolution image is reconstructed using regularization method is as follows：
f*=argmin | | g-Af | |2+λΩ(f)
Wherein Ω (f) is regularization term, and Ω is referred to as regularizing operator, and f is the high resolution image reconstructed, and A is to degrade
Operator, g are the image that observes of microlens array, and λ is referred to as regular parameter, by so can be to reconstruct high-resolution shadow
Picture.
In step 4), image is split in view of GAC models mainly stop function g using border, the g direct shadow of construction
Ring the result of segmentation；The present invention proposes the GAC model gradient flow equations based on texture gradient, as follows：
In formula, g is the nonnegative function of any monotone decreasing, and δ (x) functions can be expressed as H (x) derivative, and μ, c are normal
Number, div is divergence operator.
Present invention also offers a kind of 3-3-2 based on bionic compound eyes micro lens technology to tie up object detection system, including：It is micro-
Lens combination, control system and target detection output system；
Described microlens system includes 2 symmetrical lenticule devices, and the lenticule device can pass through lenticule
Array obtains microlens array image, then reconstructs low resolution and high-resolution image；
Described control system includes DSP main controls core cell, fpga logic control unit and graphics processing unit；
Described DSP main controls core cell is used to carry out Image Information Processing and storage；
Described fpga logic control unit is used for the signal acquisition for controlling microlens system, and is DSP main control cores
Unit provides processing data；
Described graphics processing unit is used for the low resolution three-D profile that target is carried out to above-mentioned microlens array image
Capture and high resolution 2 d staring imaging, and the target in high resolution image is accurately detected；
The target detection output system is used to export object detection results.
Further, described image processing unit includes extracting and matching feature points module, relative orientation module, front friendship
Meeting module and GAC models segmentation module,
The extracting and matching feature points module is used to carry out feature point extraction with matching to the low resolution image of acquisition；
The relative orientation module is used for the screen fraction same place in match point and carries out relative orientation；
The forward intersection module is used for the relative dimensional coordinate that match point is obtained using forward intersection measuring method；
The GAC models segmentation module is used for the segmentation identification that target is carried out using the GAC models based on texture gradient.
For the present invention due to using above technical scheme, it has advantages below：
1st, the present invention proposes the new mould of the Data acquisition and Proclssing of the 3-3-2 dimensions based on bionic compound eyes micro lens technology
Formula.The two-level resolution of the method simulated hexapod compound eye based on micro lens technology carries out low resolution three-dimensional to objective and caught
Obtain and high resolution 2 d staring imaging.Target area is imaged using microlens system first, passes through corresponding image procossing
Algorithm carries out rough detection to the three-D profile of target, after preliminary latch moving target, utilizes regularization method reconstruct target area
High resolution image, then accurately identification target.By increasing by one target three-D profile is captured using low resolution image
The step of, it can more efficiently judge that target whether in the target area, is effectively avoided to the insignificant place of redundancy image
Reason, improve the real-time treatment effeciency of system.
2nd, the present invention reconstructs low resolution image to microlens array image using the linear weighted function method of average；To low resolution
Image carries out feature extraction and matching, and relative orientation and forward intersection are reconstructed the three-D profile of target；With microlens array shadow
As based on, using the method for regularization, the high resolution image of target area is reconstructed；Using the GAC based on texture gradient
Model carries out the segmentation identification of target, can effectively identify target.
3rd, the present invention first by micro lens technology, photogrammetric, computer vision, insect bionic compound eyes correlation theory knot
Altogether, it is proposed that a set of object detection method based on insect bionic compound eyes image, the invention have pioneering, practicality.
It present invention can be widely used to target detection.
Brief description of the drawings
Fig. 1 microlens system structure charts of the present invention, wherein：1,2-lenticule device.
The single lenticule device imaging schematic diagram of Fig. 2 present invention, wherein：3-headprism, 4-microlens array.
Fig. 3 object detection method flow charts of the present invention.
Fig. 4 relative orientation schematic diagrames of the present invention.
Fig. 5 forward intersection schematic diagrames of the present invention.
Fig. 6 target three-D profile schematic diagrames of the present invention, (a) represent initial data, and (b) represents profile point cloud.
Embodiment
The present invention is described in detail with reference to the accompanying drawings and examples.
Current object detection method is utilize imaging sensor, such as objective of camera acquisition spatially two
Image is tieed up, then the image of acquisition is handled using computer.The detection of target acquisition recognition methods in traditional means
Visual field is small, the resolution ratio such as the view data of all regions of acquisition is all.The image of camera shooting is more, and imaging resolution is higher
Mean to obtain data volume is bigger, data message is more.This causes computer to carry out the required time of automatic business processing more
Long, efficiency is lower.In addition, the uncertain effectively capture target that causes to differ in high-definition picture surely of target location, that is, deposit
In substantial amounts of redundancy image.In a word, traditional object detection method can not realize that scouting, can be to substantial amounts of redundant data to scrutinizing
Insignificant processing is carried out, causes target detection real-time efficiency low.
The design of the present invention is as shown in figure 1, the microlens system includes the two lenticule devices 1 and 2 in left and right.Lenticule device
The internal structure and image-forming principle of part 1 are as shown in Figure 2.Single lenticule device includes a headprism 3, microlens array 4, light
Electric transducer (not shown).Headprism mainly carries out visual field and logical fader control to target identification region, into headprism visual field
Light will be by microlens array re-imaging, photoelectric sensor, which is substantially carried out being imaged optical signal and electric signal, to be changed.When micro-
When mirror device is to target imaging, microlens array obtains the microlens array image of target area.The present invention based on this, is led to
The low resolution image for obtaining target using the linear weighted function method of average to image is crossed, high-resolution is reconstructed by regularization method
Image.
Recognize based on more than, the present invention proposes a kind of bionic compound eyes 3-3-2 dimension target detections based on micro lens technology
Method, it can effectively avoid carrying out insignificant processing, the efficiency that raising system is handled in real time to redundancy image.
As shown in figure 3, describing the implementation process of the inventive method in figure, it comprises the following steps：
1) it is imaged.Target area is imaged using microlens system, obtains the microlens array image of target area.
Each unit image in microlens array is handled using linear weighted function averaging method, generates low point of target area
Resolution image.Because two lenticule devices in device have been fixed, so measuring the distance between two lenticule devices
Can be as the initial value of relative orientation in the 3rd step.
2) extracting and matching feature points.Feature point extraction is carried out with matching to the low resolution image of acquisition.This feature
Translation occurs between can handling two images with algorithm, rotates, the matching problem in the case of affine transformation, there is very strong
With ability.The algorithm, which extracts, is characterized in the local feature of image, and to translating, rotating, scaling, brightness change, vision become
Change, affine transformation etc. also keep preferable stability.The algorithm is determined each using topography's Gradient Features of characteristic point
The principal direction of individual characteristic point, its formula are as follows：
In formula, m (x, y) and θ (x, y) are gradient and direction of the gaussian pyramid image of current scale at (x, y) place, L
(x, y) is gray scale of the gaussian pyramid image at (x, y) place of current scale.
3) relative orientation.As shown in Figure 4, station S is taken the photograph from two1, S2When absorbing a stereogram to same bottom surface, stand
Two corresponding image rayses of any object point of body image centering all intersect at the object point, that is, corresponding image rays be present to intersecting phenomenon.
If the relative position and posture relation between two photos of holding are constant, two photos are moved integrally, rotated and changed baseline
Length, corresponding image rays to intersecting characteristic to that will not change.According to corresponding image rays to intersecting in this stereogram
Geometrical relationship, pass through the picture point m of measurement1, m2Coordinate, elements of relative orientation is sought with the method for analytical Calculation, i.e., in two figures
Overlapping region as in chooses a number of same place and carries out relative orientation, determines two lenticule devices in the relative of space
Three-dimensional coordinate, that is, calculate the elements of relative orientation of two photosWhereinIt is relative for second photo
The elements of interior orientation of first photo, (μ, ν) are the drift angle and inclination angle of baseline.Solve elements of relative orientation equation be:
Wherein,Wherein Q is to regard up and down
Difference, N1, N2For projection coefficient, (X1,Y1,Z1), (X2,Y2,Z2) it is that coordinate of the picture point in the auxiliary coordinates of image space is using most
A young waiter in a wineshop or an inn multiplies principle, can establish an equation
Elements of relative orientation can be tried to achieve by indirect adjustment
4) space intersection.As shown in figure 5, using the forward intersection method in photogrammetric, that is, utilize the interior of photo
The element of orientation, relative bearing element, the corresponding image points coordinate of stereogram calculate the relative dimensional coordinate of model.It is substantial amounts of same
Famous cake three-dimensional coordinate, that is, the three-D profile of target is determined.It is comprised the following steps that：Angular range element and baseline are calculated first
Component (BX、BY、BZ)；Calculate the direction cosines of spin matrix of the left and right photo in photogrammetric coordinate system；Picture point is calculated in picture
Coordinate (X in the auxiliary coordinates of space1、Y1、Z1) and (X2、Y2、Z2)；Calculate projection coefficient N1、N2；The three-dimensional of computation model point
Coordinate (X, Y, Z).Three-dimensional coordinate calculation formula of the model points in the auxiliary coordinates of image space is as follows：
The three-dimensional coordinate that substantial amounts of model points are obtained by the above method obtains the three-D profile of target.Fig. 6 is represented
The three-D profile of the target obtained by the above method.
5) high resolution image is reconstructed.According to the three-D profile generated in step 4), judge whether microlens system is effective
Capture target.If microlens system effectively captures target, based on microlens array image, regularization method is used
Reconstruct high resolution image.Described regularization method is as follows：Assuming that the process that degrades of image is：
Af=g
Wherein f is the high resolution image reconstructed, and A is the operator that degrades, and g is the image that microlens array observes.With
The least square solution problem, can be obtained,
f*=argmin | | g-Af | |2
Due to the nonuniqueness solved in the indirect problem, it is necessary to which adding extra prior information can just obtain approaching original image
Solution.Original optimization problem is converted into by regularization method：
f*=argmin | | g-Af | |2+λΩ(f)
Wherein Ω (f) is regularization term, and Ω is referred to as regularizing operator, and λ is referred to as regular parameter, i.e., with adjoining with former problem
Well-posed problem go to approach the solution of former problem.By so can be to reconstruct high-resolution image.
6) target in high resolution image is accurately detected using texture gradient GAC models.First, calculate bag
Minimum Area containing three-D profile.Second, the texture gradient on the yardstick of image i-th is solved using Gaussian derivative approximation to function method,
Different scale and the gradient magnitude in direction are：
G' in formulax, G'yIt is partial derivative of the Gaussian function in x and y directions respectively, Mi,θ(x, y) can be become by bi-input bi-output system
Change and obtain.3rd, initial circuit is set in the Minimum Area that determines in the first step, the condition of iteration is set, circuit carry out by
Step reduction finally detects moving target.The initial value of circuit sets as follows：
E represents to minimize a closed curve C (p) energy functional, when energy functional reaches minimum, corresponding curve
The border exactly split.GAC models mainly stop function g using border and image are split, and g construction, which directly affects, to be divided
The result cut, it is as follows set forth herein the GAC model gradient flow equations based on texture gradient based on this：
In formula, g is the nonnegative function of any monotone decreasing, and δ (x) functions can be expressed as H (x) derivative, and μ, c are normal
Number, div is divergence operator.
The various embodiments described above are merely to illustrate the present invention, every equivalent change carried out on the basis of technical solution of the present invention
Change and improve, should not exclude outside protection scope of the present invention.
Claims (9)
1. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, comprise the following steps：
1) seizure imaging is carried out to target area using the microlens system based on bionic compound eyes structure as imaging system, will shot
Each unit image in obtained microlens array image is using linear weighted function method of average reconstruct low resolution image；
2) in step 1) based on the low resolution image of reconstruct, the three-dimensional of target point is calculated using forward intersection measuring method
Coordinate, low resolution three-D profile capture is carried out to target；
If 3) effectively after capture target, data based on the microlens array image obtained by step 1) shooting, using canonical
Change method reconstructs high resolution image, and high resolution 2 d staring imaging is carried out to target area；Otherwise, mobile microlens system
And return to step 1)；The method that high resolution image is reconstructed using regularization method is as follows：
f*=arg min | | g-Af | |2+λΩ(f)
Wherein Ω (f) is regularization term, and Ω is referred to as regularizing operator, and f is the high resolution image reconstructed, and A is the operator that degrades,
G is the image that microlens array observes, λ is referred to as regular parameter；
4) after the high resolution 2 d image for obtaining target area, essence is carried out to target using the GAC models based on texture gradient
Really identification, complete target detection.
2. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, its feature exist as claimed in claim 1
In in step 1), the described linear weighted function method of average is as follows：
Wherein,
<mrow>
<msub>
<mi>p</mi>
<mi>i</mi>
</msub>
<mo>=</mo>
<mfrac>
<msub>
<mi>n</mi>
<mi>i</mi>
</msub>
<mrow>
<msub>
<mi>n</mi>
<mn>1</mn>
</msub>
<mo>+</mo>
<msub>
<mi>n</mi>
<mn>2</mn>
</msub>
<mo>+</mo>
<mo>...</mo>
<mo>+</mo>
<msub>
<mi>n</mi>
<mi>m</mi>
</msub>
</mrow>
</mfrac>
<mo>,</mo>
<mrow>
<mo>(</mo>
<mn>1</mn>
<mo>&le;</mo>
<mi>i</mi>
<mo>&le;</mo>
<mi>m</mi>
<mo>)</mo>
</mrow>
</mrow>
In formula, the pixel in unit image is ranked up from small to large according to gray value, and it is 1 to m to be numbered, gmRepresent
The gray value for the pixel that numbering is m, p in unit imageiFor gray value giCorresponding weight, niExpression gray value is gi's
Number of pixels.
3. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, its feature exist as claimed in claim 1
In step 2) specifically includes herein below：1. feature point extraction is carried out to image with matching；2. screen fraction is same in match point
Famous cake carries out relative orientation, determines the relative attitude information between lenticule device；3. according to the relative appearance between lenticule device
State information, the relative dimensional coordinate of match point is obtained using forward intersection measuring method, obtain the three-D profile of target.
4. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, its feature exist as claimed in claim 3
In determining the principal direction of each characteristic point using topography's Gradient Features of characteristic point, its formula is as follows：
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mrow>
<mi>m</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<msqrt>
<mrow>
<mo>(</mo>
<mi>L</mi>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>+</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mi>L</mi>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>-</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>+</mo>
<mo>(</mo>
<mi>L</mi>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>+</mo>
<mn>1</mn>
</mrow>
<mo>)</mo>
<mo>-</mo>
<mi>L</mi>
<msup>
<mrow>
<mo>(</mo>
<mrow>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>-</mo>
<mn>1</mn>
</mrow>
<mo>)</mo>
</mrow>
<mn>2</mn>
</msup>
<mo>)</mo>
</mrow>
</msqrt>
</mrow>
</mtd>
</mtr>
<mtr>
<mtd>
<mrow>
<mi>&theta;</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>=</mo>
<mi>arctan</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>+</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>,</mo>
<mi>y</mi>
<mo>-</mo>
<mn>1</mn>
<mo>)</mo>
</mrow>
</mrow>
<mrow>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>+</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
<mo>-</mo>
<mi>L</mi>
<mrow>
<mo>(</mo>
<mi>x</mi>
<mo>-</mo>
<mn>1</mn>
<mo>,</mo>
<mi>y</mi>
<mo>)</mo>
</mrow>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
</mrow>
</mtd>
</mtr>
</mtable>
</mfenced>
(x, y) is characterized coordinate a little in formula, m (x, y) and θ (x, y) be respectively current scale gaussian pyramid image (x,
Y) gradient at place and direction, L (x, y) are gray scale of the gaussian pyramid image at (x, y) place of current scale.
5. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, its feature exist as claimed in claim 3
In the equation for determining the relative attitude information between two lenticule devices is:
Wherein, Q=N1Y1-N2Y2-BY,Wherein Q is vertical parallax, N1, N2For
Projection coefficient, (X1,Y1,Z1), (X2,Y2,Z2) it is coordinate of the picture point in the auxiliary coordinates of image space, BX,BY,BZIt is photography base
Projection of the line on XYZ directions, d are differential signs, using the principle of least square, can be established an equation
In formula, l be indirect adjustment free term, an,bn,cn,dn,enFor the coefficient of error equation, vnFor the error of indirect adjustment
, by indirect adjustment can try to achieve elements of relative orientation (μ, ν,ω, γ), wherein (ω, γ) it is that second photo is relative
The elements of interior orientation of first photo, (μ, ν) are the drift angle and inclination angle of baseline.
6. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, its feature exist as claimed in claim 3
In the relative dimensional coordinate that match point is obtained using forward intersection measuring method is comprised the following steps that：Angle of departure side is calculated first
Bit element and baseline component (BX、BY、BZ)；Calculate orthogonal matrix of the left and right photo in photogrammetric coordinate system；Picture point is calculated to exist
Coordinate (X in the auxiliary coordinates of image space1、Y1、Z1) and (X2、Y2、Z2)；Calculate projection coefficient N1、N2；The three of computation model point
Dimension coordinate (X, Y, Z), three-dimensional coordinate calculation formula of the model points in the auxiliary coordinates of image space are as follows：
<mrow>
<mfenced open = "{" close = "">
<mtable>
<mtr>
<mtd>
<mi>X</mi>
<mo>=</mo>
<msub>
<mi>B</mi>
<mi>X</mi>
</msub>
<mo>+</mo>
<msub>
<mi>N</mi>
<mn>2</mn>
</msub>
<mi>X</mi>
</mtd>
</mtr>
<mtr>
<mtd>
<mi>Y</mi>
<mo>=</mo>
<mfrac>
<mn>1</mn>
<mn>2</mn>
</mfrac>
<mo>(</mo>
<msub>
<mi>N</mi>
<mn>1</mn>
</msub>
<msub>
<mi>Y</mi>
<mn>1</mn>
</msub>
<mo>+</mo>
<msub>
<mi>N</mi>
<mn>2</mn>
</msub>
<msub>
<mi>Y</mi>
<mn>2</mn>
</msub>
<mo>+</mo>
<msub>
<mi>B</mi>
<mi>Y</mi>
</msub>
<mo>)</mo>
</mtd>
</mtr>
<mtr>
<mtd>
<mi>Z</mi>
<mo>=</mo>
<msub>
<mi>B</mi>
<mi>Z</mi>
</msub>
<mo>+</mo>
<msub>
<mi>N</mi>
<mn>2</mn>
</msub>
<msub>
<mi>Z</mi>
<mn>2</mn>
</msub>
</mtd>
</mtr>
</mtable>
</mfenced>
<mo>.</mo>
</mrow>
7. the 3-3-2 dimension object detection methods based on bionic compound eyes micro lens technology, its feature exist as claimed in claim 1
In in step 4), accurately being identified to target using the GAC models based on texture gradient, the GAC models based on texture gradient
Gradient flow equation, it is as follows：
<mrow>
<mfrac>
<mrow>
<mo>&part;</mo>
<mi>u</mi>
</mrow>
<mrow>
<mo>&part;</mo>
<mi>t</mi>
</mrow>
</mfrac>
<mo>=</mo>
<mi>&mu;</mi>
<mo>&lsqb;</mo>
<mo>&dtri;</mo>
<mi>u</mi>
<mo>-</mo>
<mi>d</mi>
<mi>i</mi>
<mi>v</mi>
<mrow>
<mo>(</mo>
<mfrac>
<mrow>
<mo>&dtri;</mo>
<mi>u</mi>
</mrow>
<mrow>
<mo>|</mo>
<mo>&dtri;</mo>
<mi>u</mi>
<mo>|</mo>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>&delta;</mi>
<mrow>
<mo>(</mo>
<mi>u</mi>
<mo>)</mo>
</mrow>
<mo>&lsqb;</mo>
<mi>d</mi>
<mi>i</mi>
<mi>v</mi>
<mrow>
<mo>(</mo>
<mi>g</mi>
<mo>(</mo>
<mrow>
<mi>T</mi>
<mi>G</mi>
</mrow>
<mo>)</mo>
<mfrac>
<mrow>
<mo>&dtri;</mo>
<mi>u</mi>
</mrow>
<mrow>
<mo>|</mo>
<mo>&dtri;</mo>
<mi>u</mi>
<mo>|</mo>
</mrow>
</mfrac>
<mo>)</mo>
</mrow>
<mo>+</mo>
<mi>c</mi>
<mi>g</mi>
<mo>&rsqb;</mo>
</mrow>
In formula, g is the nonnegative function of any monotone decreasing, and δ (x) functions can be expressed as H (x) derivative, and μ, c are constant, div
For divergence operator.
8. the 3-3-2 dimension object detection systems based on bionic compound eyes micro lens technology, including：Microlens system, control system and
Target detection output system；
Described microlens system includes 2 symmetrical lenticule devices, and the lenticule device can pass through microlens array
Obtain microlens array image；
Described control system includes DSP main controls core cell, fpga logic control unit and graphics processing unit；
Described DSP main controls core cell is used to carry out Image Information Processing and storage；
Described fpga logic control unit is used for the signal acquisition for controlling microlens system, and is DSP main control core cells
Processing data is provided；
Described graphics processing unit is used to put down each unit image in above-mentioned microlens array image using linear weighted function
Equal method reconstruct low resolution image, and target is calculated using forward intersection measuring method based on the low resolution image of reconstruct
The three-dimensional coordinate of point is captured with carrying out the low resolution three-D profile of target, and with above-mentioned lenticule after effectively capture target
Data based on array image, high resolution image is reconstructed using regularization method, and high resolution 2 d is carried out to target area
Staring imaging, and the target in high resolution image is accurately detected, wherein, high-resolution is reconstructed using regularization method
The method of rate image is as follows：
f*=arg min | | g-Af | |2+λΩ(f)
Wherein Ω (f) is regularization term, and Ω is referred to as regularizing operator, and f is the high resolution image reconstructed, and A is the operator that degrades,
G is the image that microlens array observes, λ is referred to as regular parameter；
The target detection output system is used to export object detection results.
9. the dimension object detection systems of the 3-3-2 based on bionic compound eyes micro lens technology described in claim 8, it is characterised in that
Split module including extracting and matching feature points module, relative orientation module, forward intersection module and GAC models,
The extracting and matching feature points module is used to carry out feature point extraction with matching to the low resolution image of the acquisition；
The relative orientation module is used for the screen fraction same place in match point and carries out relative orientation；
The forward intersection module is used for the relative dimensional coordinate that match point is obtained using forward intersection measuring method；
The GAC models segmentation module is used for the segmentation identification that target is carried out using the GAC models based on texture gradient.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510732346.5A CN105352482B (en) | 2015-11-02 | 2015-11-02 | 332 dimension object detection methods and system based on bionic compound eyes micro lens technology |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201510732346.5A CN105352482B (en) | 2015-11-02 | 2015-11-02 | 332 dimension object detection methods and system based on bionic compound eyes micro lens technology |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN105352482A CN105352482A (en) | 2016-02-24 |
| CN105352482B true CN105352482B (en) | 2017-12-26 |
Family
ID=55328492
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201510732346.5A Active CN105352482B (en) | 2015-11-02 | 2015-11-02 | 332 dimension object detection methods and system based on bionic compound eyes micro lens technology |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN105352482B (en) |
Families Citing this family (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| TWI610059B (en) * | 2016-08-04 | 2018-01-01 | 緯創資通股份有限公司 | Three-dimensional measurement method and three-dimensional measurement device using the same |
| CN106339603B (en) * | 2016-09-08 | 2019-03-19 | 武汉大学 | A kind of relative orientation method based on shaft angle vector |
| CN110645956B (en) * | 2019-09-24 | 2021-07-02 | 南通大学 | Multi-channel visual ranging method imitating stereoscopic vision of insect compound eye |
| CN110989646A (en) * | 2019-12-02 | 2020-04-10 | 西安欧意特科技有限责任公司 | Compound eye imaging principle-based target space attitude processing system |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002013295A (en) * | 2000-06-30 | 2002-01-18 | Nikken Atorasu Kobe:Kk | Method for creating investigation data on deformed part in building |
| CN1932841A (en) * | 2005-10-28 | 2007-03-21 | 南京航空航天大学 | Petoscope based on bionic oculus and method thereof |
| CN102572220A (en) * | 2012-02-28 | 2012-07-11 | 北京大学 | Bionic compound eye moving object detection method adopting new 3-2-2 spatial information conversion model |
| CN103295221A (en) * | 2013-01-31 | 2013-09-11 | 河海大学 | Water surface target motion detecting method simulating compound eye visual mechanism and polarization imaging |
| CN103325088A (en) * | 2013-07-04 | 2013-09-25 | 中国科学院光电技术研究所 | Multichannel image processing method for curved compound eye imaging system |
-
2015
- 2015-11-02 CN CN201510732346.5A patent/CN105352482B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2002013295A (en) * | 2000-06-30 | 2002-01-18 | Nikken Atorasu Kobe:Kk | Method for creating investigation data on deformed part in building |
| CN1932841A (en) * | 2005-10-28 | 2007-03-21 | 南京航空航天大学 | Petoscope based on bionic oculus and method thereof |
| CN102572220A (en) * | 2012-02-28 | 2012-07-11 | 北京大学 | Bionic compound eye moving object detection method adopting new 3-2-2 spatial information conversion model |
| CN103295221A (en) * | 2013-01-31 | 2013-09-11 | 河海大学 | Water surface target motion detecting method simulating compound eye visual mechanism and polarization imaging |
| CN103325088A (en) * | 2013-07-04 | 2013-09-25 | 中国科学院光电技术研究所 | Multichannel image processing method for curved compound eye imaging system |
Non-Patent Citations (5)
| Title |
|---|
| A Method For Moving Target Capture Using 3D Profile Information Based On Bionic Compound Eye;Bo-Ren L,Lei Y,Hui-Li L;《2012 International Conference on IEEE》;20121231;全文 * |
| A Novel System for Moving Object Detection Using Bionic Compound Eyes;Sun H,Zhao H,Mooney P;《Journal of Bionic engineering》;20111231;全文 * |
| A Synchronous Imaging System for Moving-target Detection With Bionic Compound eyes;Zhang Z,Yan L,Sun H;《Imaging and Signal Processing》;20111231;全文 * |
| Imaging Mechanism of Moving Object Detection;Sun H，Tong H, Liang R;《2010 18th international conference on.IEEE》;20101231;全文 * |
| 基于332遥感信息处理模式的仿生复眼运动目标检测;百度文库;《道克巴巴 http://www.doc88.com/p-0347153926975.html》;20140319;正文第90-94页，第24页，第57-59页，第64页，第67-68页，第100-101页，图4-4-4-6，图6-11 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN105352482A (en) | 2016-02-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2021004312A1 (en) | Intelligent vehicle trajectory measurement method based on binocular stereo vision system | |
| CN109509230B (en) | SLAM method applied to multi-lens combined panoramic camera | |
| CN105678742B (en) | A kind of underwater camera scaling method | |
| CN104182982B (en) | Overall optimizing method of calibration parameter of binocular stereo vision camera | |
| Zhang et al. | A UAV-based panoramic oblique photogrammetry (POP) approach using spherical projection | |
| EP3977346A1 (en) | Simultaneous localization and mapping method, device, system and storage medium | |
| CN106485753B (en) | The method and apparatus of camera calibration for pilotless automobile | |
| CN108648240A (en) | Based on a non-overlapping visual field camera posture scaling method for cloud characteristics map registration | |
| CN111144388A (en) | Monocular image-based road sign line updating method | |
| CN106780712B (en) | Three-dimensional point cloud generation method combining laser scanning and image matching | |
| CN107481315A (en) | A kind of monocular vision three-dimensional environment method for reconstructing based on Harris SIFT BRIEF algorithms | |
| CN106485690A (en) | Cloud data based on a feature and the autoregistration fusion method of optical image | |
| CN107392964A (en) | The indoor SLAM methods combined based on indoor characteristic point and structure lines | |
| CN104034305B (en) | A kind of monocular vision is the method for location in real time | |
| CN105352482B (en) | 332 dimension object detection methods and system based on bionic compound eyes micro lens technology | |
| CN107464264A (en) | A kind of camera parameter scaling method based on GPS | |
| CN104361627B (en) | Binocular vision bituminous paving Micro texture 3-D view reconstructing method based on SIFT | |
| CN109596121B (en) | Automatic target detection and space positioning method for mobile station | |
| CN106952299B (en) | A kind of 3 d light fields Implementation Technology suitable for Intelligent mobile equipment | |
| CN103278138A (en) | Method for measuring three-dimensional position and posture of thin component with complex structure | |
| CN108830191A (en) | Based on the mobile robot SLAM method for improving EMM and ORB algorithm | |
| CN106971408A (en) | A kind of camera marking method based on space-time conversion thought | |
| CN110319772A (en) | Vision large span distance measuring method based on unmanned plane | |
| WO2021004416A1 (en) | Method and apparatus for establishing beacon map on basis of visual beacons | |
| CN109325981A (en) | Based on the microlens array type optical field camera geometrical parameter calibration method for focusing picture point |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant |