CN109191502A - A kind of method of automatic identification shell case trace - Google Patents
A kind of method of automatic identification shell case trace Download PDFInfo
- Publication number
- CN109191502A CN109191502A CN201810920607.XA CN201810920607A CN109191502A CN 109191502 A CN109191502 A CN 109191502A CN 201810920607 A CN201810920607 A CN 201810920607A CN 109191502 A CN109191502 A CN 109191502A
- Authority
- CN
- China
- Prior art keywords
- characteristic point
- shell case
- trace
- point
- case trace
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 50
- 239000013598 vector Substances 0.000 claims abstract description 20
- 230000004044 response Effects 0.000 claims description 17
- 239000011159 matrix material Substances 0.000 claims description 15
- 238000002474 experimental method Methods 0.000 claims description 5
- 230000008569 process Effects 0.000 claims description 4
- 238000012876 topography Methods 0.000 claims description 4
- 230000003321 amplification Effects 0.000 claims description 3
- 238000004458 analytical method Methods 0.000 claims description 3
- 238000006243 chemical reaction Methods 0.000 claims description 3
- 238000010276 construction Methods 0.000 claims description 3
- 230000003247 decreasing effect Effects 0.000 claims description 3
- 239000004615 ingredient Substances 0.000 claims description 3
- 238000013507 mapping Methods 0.000 claims description 3
- 238000003199 nucleic acid amplification method Methods 0.000 claims description 3
- 238000012216 screening Methods 0.000 claims description 3
- 230000009466 transformation Effects 0.000 claims description 3
- 238000011160 research Methods 0.000 description 4
- 238000005516 engineering process Methods 0.000 description 3
- 238000010586 diagram Methods 0.000 description 2
- 238000000605 extraction Methods 0.000 description 2
- 238000007689 inspection Methods 0.000 description 2
- 230000009467 reduction Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 239000012141 concentrate Substances 0.000 description 1
- 238000007334 copolymerization reaction Methods 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000004069 differentiation Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 230000008520 organization Effects 0.000 description 1
- 238000000746 purification Methods 0.000 description 1
- 238000004445 quantitative analysis Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/30—Determination of transform parameters for the alignment of images, i.e. image registration
- G06T7/33—Determination of transform parameters for the alignment of images, i.e. image registration using feature-based methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/46—Descriptors for shape, contour or point-related descriptors, e.g. scale invariant feature transform [SIFT] or bags of words [BoW]; Salient regional features
- G06V10/462—Salient features, e.g. scale invariant feature transforms [SIFT]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T2207/00—Indexing scheme for image analysis or image enhancement
- G06T2207/20—Special algorithmic details
- G06T2207/20016—Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
Abstract
The invention discloses a kind of methods of automatic identification shell case trace, including step are as follows: acquires shell case Trace Data using three-dimensional Laser Scanning Confocal Microscope and is pre-processed;Construct shell case trace difference of Gaussian pyramid;Each pixel is traversed on the pyramid and compared with 26 points around it, using Local Extremum as characteristic point;According to calculated characteristic point, its feature vector is calculated using SIFT algorithm;According to described eigenvector, the matching of shell case trace is carried out with Euclidean distance method;Characteristic point close quarters are calculated, shell case trace gross area percentage is accounted for as judging whether matched foundation using its area.Through the above way, a kind of method of automatic identification shell case trace provided by the invention, meet the quick and precisely matched requirement of shell case trace, it can not only realize the registration of a large amount of shell case traces, and it proposes and judges the whether matched qualitative criteria of shell case trace, clue to solve the case and Evidence in Litigation can be provided for gun-related case, physical resources and financial resources are greatly saved.
Description
Technical field
The present invention relates to bullet marks recognitions to identify field, more particularly to a kind of method of automatic identification shell case trace.
Background technique
The usual property of gun-related case is severe, very harmful, it is necessary to track down and prevent as early as possible, to ensure social stability, the public
Safety.It is the important technical of gun-related case of investigating and prosecuting that shoot mark, which is examined, can provide clue to solve the case for public security and judicial department and tell
Dispute evidence.Traditional shoot mark method of inspection mainly relies on shoot mark reviewer, passes through the trace on micro- sem observation body surface
Details realizes the comparison of bullet, and time-consuming for whole process, heavy workload and subjectivity are strong.With the extensive use of artificial intelligence,
Area of computer aided shoot mark automatic identification technology is rapidly developed, and previous artificial eye observation is compared, and shoot mark determination rates are promoted
Obviously, identification process is also more scientific objective.
Since the nineties in last century, the states such as Canada, Germany, Israel have successively developed the inspection of bullet trace automatic comparison
Cable system, such as IBIS, EVOFINDER, BALSCAN and ALIAS.2000, the Ministry of Public Security of China also voluntarily had developed a set of bullet
Trace computer recognition system.In addition to this, numerous universities and research institution are also in the research for being engaged in this field.In fact,
Shell case trace automatic identification is concentrated mainly on two key technologies: high-precision shell case trace acquisition and efficient shoot mark compare other side
Method.Currently, having abandoned CCD substantially as the mainstream identifying system of representative using IBIS images acquisition method, it is burnt to be all made of three-dimensional copolymerization
Microscope obtains 3 d surface topography, can really reflect that bullet indication character is distributed.For shoot mark comparison method, the overwhelming majority is ground
Study carefully, only concentrates on target comparison, i.e., only quantify and ambiguity, and substantially using cross-correlation coefficient as reference.IBIS etc. is
The strategy of system is also to provide height according to cross-correlation coefficient to sort, then by manually carrying out secondary judgement.In fact, computer bullet
The final purpose of trace automatic identification is not only object comparison certainly, but obtains qualitative conclusions based on quantitative analysis.It can be said that mesh
The research of preceding shell case trace matching qualitative method still quite lacks, and research deeply, is not tested also insufficient.How to qualitatively judge
Or not that shell case trace matches it will also become the inexorable trend of shell case trace automatic identification field development from now on.
Summary of the invention
The present invention provides a kind of method of automatic identification shell case trace, can not only realize the quantitative comparison of shell case trace,
And it can be realized the qualitative judgement whether matching of shell case trace.
In order to solve the above technical problems, one technical scheme adopted by the invention is that: a kind of automatic identification shell case trace is provided
The method of mark, it is characterised in that the following steps are included:
A kind of method of automatic identification shell case trace, comprising the following steps:
Step (1) acquires shell case Trace Data using three-dimensional Laser Scanning Confocal Microscope and pre-processes to it;Construct shell case
Trace difference of Gaussian pyramid is to form multiscale space;
Each pixel and compared with 26 points around it in scale space described in step (2), the search step (1)
Determine characteristic point;
Step (3), according to characteristic point in the step (2), calculate its feature vector using SIFT algorithm;
Step (4), according to feature vector in the step (3), utilize euclidean distance method to carry out the matching of shell case trace;
Step (5) calculates Relatively centralized characteristic point according to feature vector in the step (4);
Step (6), the Relatively centralized characteristic point according to the step (5) calculate characteristic point close quarters area;
It is total that step (7), the calculating shell case trace gross area and step (6) the characteristic point close quarters area account for shell case trace
Area percentage proposes standard whether judgement matching.
A kind of method of automatic identification shell case trace, it is characterised in that the following steps are included:
(1) shell case Trace Data is acquired using three-dimensional Laser Scanning Confocal Microscope and pre-processed;
(2) construction difference of Gaussian pyramid is at multiscale space,
(3) it searches for each pixel in difference of Gaussian pyramid and determines characteristic point compared with 26 points around it, specifically
Steps are as follows:
Each pixel on difference of Gaussian pyramid is traversed, and with it with 8 consecutive points of scale and neighbouring scale 9
× 2 totally 26 points compare, if maximum value or minimum value point, i.e., temporarily regard as characteristic point;
(4) curvature is calculated at the step (3) characteristic point to reject unstable skirt response point by Hessian matrix;
(5) characteristic point determined according to the step (3), calculates its feature vector using SIFT algorithm;
(6) according to the feature vector acquired, the matching of shell case trace is carried out using euclidean distance method;
(7) according to matching result at the beginning of SIFT algorithm shell case trace, most of erroneous matching is rejected using RANSAC algorithm;
(8) Relatively centralized characteristic point is calculated;
(9) based on Relatively centralized characteristic point, the algorithm for determining characteristic point close quarters is proposed;
(10) it needing to calculate the shell case trace gross area after acquiring characteristic point close quarters area, such trace shape is annular,
Area, that is, great circle of trace subtracts the area of roundlet.
Specific step is as follows for the step (1):
1) shell case Trace Data is acquired using three-dimensional Laser Scanning Confocal Microscope, can reflects its 3D surface appearance feature;
2) it is filtered using low pass spline filter to weaken percent ripple ingredient;
3) it is cut with appropriate threshold value ± Tr at the top and bottom of trace, beyond partially clipping;
4) shell case trace three dimensional topography is integrally added into Tr, makes to move to zero or more in the texture whole of bottom;
5) Trace Data is whole multiplied by amplification coefficient Am, makes data conversion to common image intensity range (0~255).
Specific step is as follows for the step (2):
1) scale space L (x, y, σ)=G (x, y, σ) * I (x, y) of a width two dimensional image is defined, wherein * is indicated in x and y
Convolution algorithm on direction, G (x, y, σ)=1/2 π σ2·exp((x2+y2)/2σ2), G (x, y, σ) is a changeable scale Gaussian function
Number, (x, y) is space coordinate, and σ is scale coordinate;
2) Gaussian Blur of different scale is done to shell case trace and original trace constantly down-sampled is obtained into a series of sizes
Different image, these images are descending, tower structure is constituted from bottom to top;This structure one is divided into O group, and every group S layers,
Form gaussian pyramid;
3) shell case mark image adjacent with group in gaussian pyramid is subtracted each other two-by-two, obtains difference of Gaussian pyramid.
Specific step is as follows for the step (4):
1) image can generate stronger skirt response in difference of Gaussian pyramid, need to reject unstable skirt response
Point has biggish principal curvatures in the direction across edge, and has lesser principal curvatures in the direction of vertical edge;Principal curvatures can
To be found out by 2 × 2 Hessian matrix H:
In formula, D is differential operator;
2) enabling α is maximum eigenvalue, and β is minimal eigenvalue;The sum that them are calculated by the mark of H-matrix, passes through H-matrix
Their product of determinant computation: Tr (H)=Dxx+Dyy=alpha+beta
Det (H)=DxxDyy-(Dxy)2=α β
3) α=γ β is enabled, then is had
FormulaValue it is minimum when two characteristic values are equal, value two characteristic value ratios of bigger explanation are bigger, into
It is bigger in the gradient value of a direction that one step illustrates this characteristic point, and the gradient value in other direction is smaller, and here it is edges to ring
The case where answering;So in order to reject skirt response point, it is only necessary to make?;Set γ value;In satisfaction
The characteristic point of formula retains, and ungratified characteristic point is rejected.
Specific step is as follows for the step (5):
It 1) is each characteristic point assigned direction parameter using the gradient direction distribution characteristic of characteristic point neighborhood territory pixel,
Above formula is respectively the modulus value of gradient and direction at (x, y);
2) after the gradient for completing characteristic point calculates, the gradient direction and amplitude of pixel in statistics with histogram neighborhood are used.
0 °~360 ° of range is divided into 36 columns by gradient orientation histogram, and every 10 ° are a column.Finally take histogram peak direction
As characteristic point principal direction, other reach the direction of peak value 80% as auxiliary direction;
3) gradient of each pixel in 4 × 4=16 window around characteristic point is calculated, and separate using the reduction of Gauss decreasing function
The weight at center ultimately forms a 128 dimensional feature description vectors.
Specific step is as follows for the step (6):
1) some characteristic point in trace sample is taken, itself and feature description vectors Euclidean distance in trace sample subject to registration are found out
Nearest the first two characteristic point;
2) in the two characteristic points, if minimum distance is less than some proportion threshold value divided by secondary short distance, receive this
A pair of of match point;
3) this proportion threshold value is reduced, SIFT match point number can be reduced, but more stable;By lots of comparing experiments,
Applied to the matched threshold value of shell case trace preferably 0.8.
Specific step is as follows for the step (7):
1) 4 pairs of characteristic points pair are extracted from matching characteristic point centering at random, calculates mapping matrix H and transformation model M;
2) error of remaining characteristic point pair with model M is calculated, if error is less than threshold value, characteristic point is stored in point set I;
3) 4 pairs of characteristic points pair constantly are randomly selected, until point set I element number is maximum;
4) characteristic point of non-point set I is rejected.
Specific step is as follows for the step (8):
The matching characteristic point determined by SIFT and RANSAC combinational algorithm be distributed on entire shell case trace ring it is more dispersed,
The characteristic point of those Relatively centralizeds can be found out and determine characteristic point close quarters.By experimental analysis, following search rule are established
Then: the feature point number for being less than T pixel apart from it around some characteristic point is more than or equal to 6, that is, assert that this characteristic point is opposite
Characteristic point is concentrated, wherein preferably 18 or 20 T;After the screening of above-mentioned algorithm, isolated characteristic point is effectively rejected relatively;It is remaining special
Sign point Relatively centralized, some characteristic point close quarters are constituted on shell case trace ring and distribution is relatively uniform.
Specific step is as follows for the algorithm of the step (9):
1) a point a optionally in Relatively centralized characteristic point, finds with it apart from nearest point b.If two o'clock distance is less than
They are then stored in set A by T;
2) using b as basic point, same operation is repeated, is found with it apart from nearest next point c, distance is less than T, c
It is stored in set A;Until two o'clock distance is greater than T;
3) if element A number is more than or equal to 5, set A becomes close quarters feature point set;Each feature point set is surrounded
Minimal convex polygon can regard as characteristic point close quarters.
The beneficial effects of the present invention are: the method for automatic identification shell case trace of the invention, it is quickly quasi- to meet shell case trace
True matched requirement, can not only realize the registration of a large amount of shell case traces, and propose and judge whether shell case trace is matched
Qualitative criteria can provide clue to solve the case and Evidence in Litigation for gun-related case, physical resources and financial resources are greatly saved, be known automatically by shoot mark
The attention and concern of other system manufacturer and public security material evidence appraisal organization.
Detailed description of the invention
Fig. 1 is the flow chart of one preferred embodiment of method of automatic identification shell case trace of the present invention;
Fig. 2 a is that characteristic point distribution map before skirt response is rejected in the method for automatic identification shell case trace of the present invention;
Fig. 2 b is that characteristic point distribution map after skirt response is rejected in the method for automatic identification shell case trace of the present invention;
Fig. 3 a is that RANSAC algorithm matches line before rejecting erroneous matching in the method for automatic identification shell case trace of the present invention
Figure;
Fig. 3 b is that RANSAC algorithm matches line after rejecting erroneous matching in the method for automatic identification shell case trace of the present invention
Figure;
Fig. 4 is characteristic point distribution before and after extracting Relatively centralized characteristic point in the method for automatic identification shell case trace of the present invention
Figure;
Fig. 5 is characteristic point close quarters area schematic diagram in the method for automatic identification shell case trace of the present invention;
Fig. 6 is shell case trace gross area schematic diagram in the method for automatic identification shell case trace of the present invention.
Specific embodiment
The preferred embodiments of the present invention will be described in detail with reference to the accompanying drawing, so that advantages and features of the invention energy
It is easier to be readily appreciated by one skilled in the art, so as to make a clearer definition of the protection scope of the present invention.
Such as Fig. 1 to Fig. 6,
Referring to Fig. 1, the present invention provides a kind of method of automatic identification shell case trace, including step are as follows:
(1) shell case Trace Data is acquired using three-dimensional Laser Scanning Confocal Microscope and pre-processed, the specific steps are as follows:
1) shell case Trace Data is acquired using three-dimensional Laser Scanning Confocal Microscope, can reflects its 3D surface appearance feature;
2) it is filtered using low pass spline filter to weaken percent ripple ingredient;
3) it is cut with appropriate threshold value ± Tr at the top and bottom of trace, beyond partially clipping;
4) shell case trace three dimensional topography is integrally added into Tr, makes to move to zero or more in the texture whole of bottom;
5) Trace Data is whole multiplied by amplification coefficient Am, makes data conversion to common image intensity range (0~255).
(2) construction difference of Gaussian pyramid is at multiscale space, the specific steps are as follows:
1) scale space L (x, y, σ)=G (x, y, σ) * I (x, y) of a width two dimensional image is defined, wherein * is indicated in x and y
Convolution algorithm on direction, G (x, y, σ)=1/2 π σ2·exp((x2+y2)/2σ2), G (x, y, σ) is a changeable scale Gaussian function
Number, (x, y) is space coordinate, and σ is scale coordinate;
2) Gaussian Blur of different scale is done to shell case trace and original trace constantly down-sampled is obtained into a series of sizes
Different image, these images are descending, tower structure is constituted from bottom to top.This structure one is divided into O group, and every group S layers,
Form gaussian pyramid;
3) shell case mark image adjacent with group in gaussian pyramid is subtracted each other two-by-two, obtains difference of Gaussian pyramid;
(3) it searches for each pixel in difference of Gaussian pyramid and determines characteristic point compared with 26 points around it, specifically
Step
It is as follows:
Each pixel on difference of Gaussian pyramid is traversed, and with it with 8 consecutive points of scale and neighbouring scale 9
× 2 totally 26 points compare, if maximum value or minimum value point, i.e., temporarily regard as characteristic point;
(4) by curvature at Hessian matrix calculating characteristic point to reject unstable skirt response point, the specific steps are as follows:
1) image can generate stronger skirt response in difference of Gaussian pyramid, need to reject unstable skirt response
Point has biggish principal curvatures in the direction across edge, and has lesser principal curvatures in the direction of vertical edge.Principal curvatures can
To be found out by 2 × 2 Hessian matrix H:
In formula, D is differential operator.
2) enabling α is maximum eigenvalue, and β is minimal eigenvalue.The sum that them are calculated by the mark of H-matrix, passes through H-matrix
Their product of determinant computation: Tr (H)=Dxx+Dyy=alpha+beta
Det (H)=DxxDyy-(Dxy)2=α β
3) α=γ β is enabled, then is had
FormulaValue it is minimum when two characteristic values are equal, value two characteristic value ratios of bigger explanation are bigger, into
It is bigger in the gradient value of a direction that one step illustrates this characteristic point, and the gradient value in other direction is smaller, and here it is edges to ring
The case where answering.So in order to reject skirt response point, it is only necessary to make?.Set an appropriate γ value.It is full
The characteristic point of sufficient above formula retains, and ungratified characteristic point is rejected.Fig. 2 is that a sample embodiment rejects characteristic point before and after skirt response
Distribution.By can see in Fig. 2 (a), characteristic point quantity is more, is distributed on trace more dispersed.Minority is located in Fig. 2 (b)
The characteristic point of edge is effectively rejected, remaining characteristic point generally within the convex peak or trench of trace at.One side image border
On point be difficult to position, have positioning ambiguousness, another aspect marginal point becomes unstable vulnerable to noise jamming, therefore picks
Except edge response point can make registration result more stable more reliable;
(5) according to determining characteristic point, its feature vector is calculated using SIFT algorithm, the specific steps are as follows:
It 1) is each characteristic point assigned direction parameter using the gradient direction distribution characteristic of characteristic point neighborhood territory pixel,
Above formula is respectively the modulus value of gradient and direction at (x, y);
2) after the gradient for completing characteristic point calculates, the gradient direction and amplitude of pixel in statistics with histogram neighborhood are used.
0 °~360 ° of range is divided into 36 columns by gradient orientation histogram, and every 10 ° are a column.Finally take histogram peak direction
As characteristic point principal direction, other reach the direction of peak value 80% as auxiliary direction;
3) gradient of each pixel in 4 × 4=16 window around characteristic point is calculated, and separate using the reduction of Gauss decreasing function
The weight at center ultimately forms a 128 dimensional feature description vectors;
(6) according to the feature vector acquired, the matching of shell case trace is carried out using euclidean distance method, the specific steps are as follows:
1) some characteristic point in trace sample is taken, itself and feature description vectors Euclidean distance in trace sample subject to registration are found out
Nearest the first two characteristic point;
2) in the two characteristic points, if minimum distance is less than some proportion threshold value divided by secondary short distance, receive this
A pair of of match point;
3) this proportion threshold value is reduced, SIFT match point number can be reduced, but more stable.By lots of comparing experiments,
Applied to the matched threshold value of shell case trace preferably 0.8;
(7) according to matching result at the beginning of SIFT algorithm shell case trace, most of erroneous matching, tool are rejected using RANSAC algorithm
Steps are as follows for body:
1) 4 pairs of characteristic points pair are extracted from matching characteristic point centering at random, calculates mapping matrix H and transformation model M;
2) error of remaining characteristic point pair with model M is calculated, if error is less than threshold value, characteristic point is stored in point set I;
3) 4 pairs of characteristic points pair constantly are randomly selected, until point set I element number is maximum;
4) characteristic point of non-point set I is rejected;
Fig. 3 is embodiment sample Feature Points Matching line graph after the purification of RANSAC algorithm.As can be seen that the overwhelming majority
For erroneous matching to effectively being rejected, remaining matching double points line is substantially parallel, and registration effect is promoted obvious;
(8) Relatively centralized characteristic point is calculated, the specific steps are as follows:
The matching characteristic point determined by SIFT and RANSAC combinational algorithm be distributed on entire shell case trace ring it is more dispersed,
The characteristic point of those Relatively centralizeds can be found out and determine characteristic point close quarters.By experimental analysis, following search rule are established
Then: the feature point number for being less than T pixel apart from it around some characteristic point is more than or equal to 6, that is, assert that this characteristic point is opposite
Characteristic point is concentrated, wherein preferably 18 or 20 T.Fig. 4 is characteristic point distribution before and after an embodiment sample extraction Relatively centralized characteristic point.
It can be seen that former characteristic point distribution is more dispersed, after the screening of above-mentioned algorithm, isolated characteristic point is effectively rejected relatively.It is surplus
Remaining characteristic point Relatively centralized, constitutes some characteristic point close quarters on shell case trace ring and distribution is relatively uniform;
(9) based on Relatively centralized characteristic point, the algorithm for determining characteristic point close quarters is proposed.Algorithm specific steps are such as
Under:
1) a point a optionally in Relatively centralized characteristic point, finds with it apart from nearest point b.If two o'clock distance is less than
They are then stored in set A by T;
2) using b as basic point, same operation is repeated, next point c (distance is less than T) recently with its distance is found, c
It is stored in set A;Until two o'clock distance is greater than T;
3) if element A number is more than or equal to 5, set A becomes close quarters feature point set.Each feature point set is surrounded
Minimal convex polygon can regard as characteristic point close quarters.Multiple close quarters can be denoted as S1, S2, S3 ... ...;
Characteristic point close quarters are assert according to the above method, calculate each region area.Fig. 5 is that an embodiment sample extraction is special
Sign point a close quarters, calculate each close quarters area S1 to S6 be respectively 1.6553e+03,1.3095e+03,652.8691,
2.4021e+03,2.5778e+03 and 1.8507e+03, in the hope of characteristic point close quarters the gross area be S=
10.4483e+03;
(10) it needing to calculate the shell case trace gross area after acquiring characteristic point close quarters area, such trace shape is annular,
Area, that is, great circle of trace subtracts the area of roundlet, Fig. 6, that is, bullet trace gross area.It can be in the hope of the area S of traceAlways=
68.7663e+03.Close quarters area account for percentage η=(10.4483e+03)/(68.7663e+03) of the trace gross area=
15.1939%.
Table 1 is 10 groups of indication character point close quarters areas, the trace gross area and area accounting.As can be seen that experiment
Object shell case trace area maximum 71.2917e+03, minimum 55.2862e+03, characteristic point close quarters area account for the total face of trace
Product percentage η maximum 19.1128%, minimum 12.8440%, area accounting is substantially between 10%~20%.Based on experiment
As a result, we can propose a judgement percentage threshold: characteristic point close quarters area accounts for trace gross area percentage and is more than
Threshold value then assert that two width shell case traces match, otherwise regards as mismatching.In fact, known unmatched trace is obtained
Characteristic point close quarters are substantially not present to usually not more than 4 pairs in characteristic point.It follows that being accounted for characteristic point close quarters
Bullet sole mark mark gross area percentage is foundation, whether can effectively determining the matching of shell case trace.
For the problems such as trace quantity is big, type is more, manual identified is difficult in shell case trace automatic identification, the present invention is proposed
A kind of shell case trace method for registering based on SIFT algorithm realizes the quantitative comparison of trace, meanwhile, propose a kind of trace of experience
Mark matches the qualitative differentiation that determination method realizes shell case trace.
SIFT searches for Local Extremum in difference of Gaussian pyramid and generates as characteristic point, and according to its neighborhood territory pixel gradient
128 dimensional feature description vectors realize scale and the description of invariable rotary feature of trace.Initial is searched for by Euclidean distance method
It is purified with point pair, and using RANSAC consistency algorithm, rejects error hiding, realize the registration of shell case trace substantially.
Based on existing matching characteristic point pair, determines characteristic point close quarters and calculate its area.With close quarters area
Accounting is as the index qualitatively judged in the trace gross area.Algorithm matches shell case trace by 10 groups 20 two-by-two and is tested
Card, the experimental results showed that, area accounting is substantially between 10~20%.Therefore, if area accounting is greater than a certain given threshold,
It can assert that two traces match, it is on the contrary then regard as mismatching.In conclusion shell case may be implemented by characteristic point compact district domain method
The qualitative judgement that trace compares.
The above description is only an embodiment of the present invention, is not intended to limit the scope of the invention, all to utilize this hair
Equivalent structure or equivalent flow shift made by bright specification and accompanying drawing content is applied directly or indirectly in other technologies neck
Domain is included within the scope of the present invention.
Claims (10)
1. a kind of method of automatic identification shell case trace, which comprises the following steps:
Step (1) acquires shell case Trace Data using three-dimensional Laser Scanning Confocal Microscope and pre-processes to it;Construct shell case trace
Difference of Gaussian pyramid is to form multiscale space;
Step (2) is searched for each pixel in scale space described in the step (1) and is determined compared with 26 points around it
Characteristic point;
Step (3), according to characteristic point in the step (2), calculate its feature vector using SIFT algorithm;
Step (4), according to feature vector in the step (3), utilize euclidean distance method to carry out the matching of shell case trace;
Step (5) calculates Relatively centralized characteristic point according to feature vector in the step (4);
Step (6), the Relatively centralized characteristic point according to the step (5) calculate characteristic point close quarters area;
Step (7), the calculating shell case trace gross area and step (6) the characteristic point close quarters area account for the shell case trace gross area
Percentage proposes standard whether judgement matching.
2. a kind of method of automatic identification shell case trace, it is characterised in that the following steps are included:
(1) shell case Trace Data is acquired using three-dimensional Laser Scanning Confocal Microscope and pre-processed;
(2) construction difference of Gaussian pyramid is at multiscale space,
(3) it searches for each pixel in difference of Gaussian pyramid and determines characteristic point, specific steps compared with 26 points around it
It is as follows: each pixel on traversal difference of Gaussian pyramid, and with it with 8 consecutive points of scale and neighbouring scale 9 × 2
Totally 26 points compare, and if maximum value or minimum value point, i.e., temporarily regard as characteristic point;
(4) curvature is calculated at the step (3) characteristic point to reject unstable skirt response point by Hessian matrix;
(5) characteristic point determined according to the step (3), calculates its feature vector using SIFT algorithm;
(6) according to the feature vector acquired, the matching of shell case trace is carried out using euclidean distance method;
(7) according to matching result at the beginning of SIFT algorithm shell case trace, most of erroneous matching is rejected using RANSAC algorithm;
(8) Relatively centralized characteristic point is calculated;
(9) based on Relatively centralized characteristic point, the algorithm for determining characteristic point close quarters is proposed;
(10) it needs to calculate the shell case trace gross area after acquiring characteristic point close quarters area, such trace shape is annular, trace
Area, that is, great circle subtract the area of roundlet.
3. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (1) it is specific
Steps are as follows:
1) shell case Trace Data is acquired using three-dimensional Laser Scanning Confocal Microscope, can reflects its 3D surface appearance feature;
2) it is filtered using low pass spline filter to weaken percent ripple ingredient;
3) it is cut with appropriate threshold value ± Tr at the top and bottom of trace, beyond partially clipping;
4) shell case trace three dimensional topography is integrally added into Tr, makes to move to zero or more in the texture whole of bottom;
5) Trace Data is whole multiplied by amplification coefficient Am, makes data conversion to common image intensity range (0~255).
4. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (2) it is specific
Steps are as follows:
1) scale space L (x, y, σ)=G (x, y, σ) * I (x, y) of a width two dimensional image is defined, wherein * is indicated in the direction x and y
On convolution algorithm, G (x, y, σ)=1/2 π σ2·exp((x2+y2)/2σ2), G (x, y, σ) is a changeable scale Gaussian function,
(x, y) is space coordinate, and σ is scale coordinate;
2) to shell case trace do different scale Gaussian Blur and by original trace constantly it is down-sampled obtain it is a series of not of uniform size
Image, these images are descending, constitute tower structure from bottom to top;This structure one is divided into O group, every group S layers, i.e. shape
At gaussian pyramid;
3) shell case mark image adjacent with group in gaussian pyramid is subtracted each other two-by-two, obtains difference of Gaussian pyramid.
5. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (4) it is specific
Steps are as follows:
1) image can generate stronger skirt response in difference of Gaussian pyramid, need to reject unstable skirt response point,
There is biggish principal curvatures in the direction across edge, and has lesser principal curvatures in the direction of vertical edge;Principal curvatures can lead to
2 × 2 Hessian matrix H is crossed to find out:
In formula, D is differential operator;
2) enabling α is maximum eigenvalue, and β is minimal eigenvalue;The sum that them are calculated by the mark of H-matrix passes through the ranks of H-matrix
Formula calculates their product: Tr (H)=Dxx+Dyy=alpha+beta
Det (H)=DxxDyy-(Dxy)2=α β
3) α=γ β is enabled, then is had
FormulaValue it is minimum when two characteristic values are equal, value two characteristic value ratios of bigger explanation are bigger, further
It is bigger in the gradient value of a direction to illustrate this characteristic point, and the gradient value in other direction is smaller, here it is skirt responses
Situation;So in order to reject skirt response point, it is only necessary to make?;Set γ value;Meet the spy of above formula
Sign point retains, and ungratified characteristic point is rejected.
6. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (5) it is specific
Steps are as follows:
It 1) is each characteristic point assigned direction parameter using the gradient direction distribution characteristic of characteristic point neighborhood territory pixel,
Above formula is respectively the modulus value of gradient and direction at (x, y);
2) after the gradient for completing characteristic point calculates, the gradient direction and amplitude of pixel in statistics with histogram neighborhood are used.Gradient
0 °~360 ° of range is divided into 36 columns by direction histogram, and every 10 ° are a column.Finally take the conduct of histogram peak direction
Characteristic point principal direction, other reach the direction of peak value 80% as auxiliary direction;
3) gradient of each pixel in 4 × 4=16 window around characteristic point is calculated, and is reduced using Gauss decreasing function far from center
Weight, ultimately form a 128 dimensional feature description vectors.
7. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (6) it is specific
Steps are as follows:
1) some characteristic point in trace sample is taken, it is nearest with feature description vectors Euclidean distance in trace sample subject to registration to find out it
The first two characteristic point;
2) in the two characteristic points, if minimum distance is less than some proportion threshold value divided by secondary short distance, receive this pair
Match point;
3) this proportion threshold value is reduced, SIFT match point number can be reduced, but more stable;By lots of comparing experiments, application
In the matched threshold value of shell case trace preferably 0.8.
8. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (7) it is specific
Steps are as follows:
1) 4 pairs of characteristic points pair are extracted from matching characteristic point centering at random, calculates mapping matrix H and transformation model M;
2) error of remaining characteristic point pair with model M is calculated, if error is less than threshold value, characteristic point is stored in point set I;
3) 4 pairs of characteristic points pair constantly are randomly selected, until point set I element number is maximum;
4) characteristic point of non-point set I is rejected.
9. the method for automatic identification shell case trace according to claim 2, which is characterized in that the step (8) it is specific
Steps are as follows:
The matching characteristic point determined by SIFT and RANSAC combinational algorithm be distributed on entire shell case trace ring it is more dispersed, can be with
It finds out the characteristic point of those Relatively centralizeds and determines characteristic point close quarters.By experimental analysis, following search rule is established: certain
The feature point number for being less than T pixel apart from it around a characteristic point is more than or equal to 6, that is, assert that this characteristic point is Relatively centralized spy
Point is levied, wherein preferably 18 or 20 T;After the screening of above-mentioned algorithm, isolated characteristic point is effectively rejected relatively;Residue character point phase
To concentration, some characteristic point close quarters are constituted on shell case trace ring and distribution is relatively uniform.
10. the method for automatic identification shell case trace according to claim 2, which is characterized in that the algorithm of the step (9)
Specific step is as follows:
1) a point a optionally in Relatively centralized characteristic point, finds with it apart from nearest point b.If two o'clock distance is less than T,
They are stored in set A;
2) using b as basic point, same operation is repeated, is found and is less than T with its nearest next point c of distance, distance, c is also stored in
Set A;Until two o'clock distance is greater than T;
3) if element A number is more than or equal to 5, set A becomes close quarters feature point set;Each feature point set is surrounded most
Small convex polygon can regard as characteristic point close quarters.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810920607.XA CN109191502B (en) | 2018-08-14 | 2018-08-14 | Method for automatically identifying cartridge case trace |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810920607.XA CN109191502B (en) | 2018-08-14 | 2018-08-14 | Method for automatically identifying cartridge case trace |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109191502A true CN109191502A (en) | 2019-01-11 |
CN109191502B CN109191502B (en) | 2021-08-24 |
Family
ID=64921376
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810920607.XA Active CN109191502B (en) | 2018-08-14 | 2018-08-14 | Method for automatically identifying cartridge case trace |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109191502B (en) |
Cited By (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109802655A (en) * | 2019-02-25 | 2019-05-24 | 哈尔滨工业大学 | Isotropism two dimension high-order spline filtering method for surface metrology |
CN110443233A (en) * | 2019-09-16 | 2019-11-12 | 上海市刑事科学技术研究院 | Data processing method, device, system and the electronic equipment that criminal investigation material evidence is saved from damage |
CN110555389A (en) * | 2019-08-09 | 2019-12-10 | 南京工业大学 | bullet line-bore trace identification method based on ridgelet transformation and rotation matching |
CN111442738A (en) * | 2020-03-23 | 2020-07-24 | 四川大学 | Device and method for acquiring three-dimensional trace characteristics of primer of cartridge case |
CN112085045A (en) * | 2020-04-07 | 2020-12-15 | 昆明理工大学 | Linear trace similarity matching algorithm based on improved longest common substring |
CN113624164A (en) * | 2021-08-06 | 2021-11-09 | 吉林省计量科学研究院 | Low-cost portable axle curved surface micro-scratch non-contact detection device and method |
CN113744238A (en) * | 2021-09-01 | 2021-12-03 | 南京工业大学 | Method for establishing bullet trace database |
CN116311628A (en) * | 2023-05-23 | 2023-06-23 | 合肥智辉空间科技有限责任公司 | Method and system for detecting safety performance of intelligent door lock |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103150728A (en) * | 2013-03-04 | 2013-06-12 | 北京邮电大学 | Vision positioning method in dynamic environment |
CN104123546A (en) * | 2014-07-25 | 2014-10-29 | 黑龙江省科学院自动化研究所 | Multi-dimensional feature extraction based bullet trace comparison method |
CN105069089A (en) * | 2015-08-04 | 2015-11-18 | 小米科技有限责任公司 | Picture detection method and device |
CN105139031A (en) * | 2015-08-21 | 2015-12-09 | 天津中科智能识别产业技术研究院有限公司 | Data processing method based on subspace clustering |
CN105654507A (en) * | 2015-12-24 | 2016-06-08 | 北京航天测控技术有限公司 | Vehicle outer contour dimension measuring method based on image dynamic feature tracking |
CN106846242A (en) * | 2015-12-07 | 2017-06-13 | 北京航天长峰科技工业集团有限公司 | The less efficient image method for registering in overlapping region is directed in a kind of video-splicing |
CN107123164A (en) * | 2017-03-14 | 2017-09-01 | 华南理工大学 | Keep the three-dimensional rebuilding method and system of sharp features |
-
2018
- 2018-08-14 CN CN201810920607.XA patent/CN109191502B/en active Active
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103150728A (en) * | 2013-03-04 | 2013-06-12 | 北京邮电大学 | Vision positioning method in dynamic environment |
CN104123546A (en) * | 2014-07-25 | 2014-10-29 | 黑龙江省科学院自动化研究所 | Multi-dimensional feature extraction based bullet trace comparison method |
CN105069089A (en) * | 2015-08-04 | 2015-11-18 | 小米科技有限责任公司 | Picture detection method and device |
CN105139031A (en) * | 2015-08-21 | 2015-12-09 | 天津中科智能识别产业技术研究院有限公司 | Data processing method based on subspace clustering |
CN106846242A (en) * | 2015-12-07 | 2017-06-13 | 北京航天长峰科技工业集团有限公司 | The less efficient image method for registering in overlapping region is directed in a kind of video-splicing |
CN105654507A (en) * | 2015-12-24 | 2016-06-08 | 北京航天测控技术有限公司 | Vehicle outer contour dimension measuring method based on image dynamic feature tracking |
CN107123164A (en) * | 2017-03-14 | 2017-09-01 | 华南理工大学 | Keep the three-dimensional rebuilding method and system of sharp features |
Non-Patent Citations (4)
Title |
---|
DAVID G. LOWE: "Distinctive Image Features from Scale-Invariant Keypoints", 《INTERNATIONAL JOURNAL OF COMPUTER VISION》 * |
HAO ZHANG ET AL.: "Pilot study of feature-based algorithm for breech face comparison", 《FORENSIC SCIENCE INTERNATIONAL》 * |
孙伟晔: "基于SIFT算法的快速图像配准技术", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
陈广居 等: "基于局部显著特征的快速图像配准方法", 《计算机应用研究》 * |
Cited By (13)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109802655B (en) * | 2019-02-25 | 2019-11-05 | 哈尔滨工业大学 | Isotropism two dimension high-order spline filtering method for surface metrology |
CN109802655A (en) * | 2019-02-25 | 2019-05-24 | 哈尔滨工业大学 | Isotropism two dimension high-order spline filtering method for surface metrology |
CN110555389B (en) * | 2019-08-09 | 2022-02-25 | 南京工业大学 | Bullet line-bore trace identification method based on ridgelet transformation and rotation matching |
CN110555389A (en) * | 2019-08-09 | 2019-12-10 | 南京工业大学 | bullet line-bore trace identification method based on ridgelet transformation and rotation matching |
CN110443233A (en) * | 2019-09-16 | 2019-11-12 | 上海市刑事科学技术研究院 | Data processing method, device, system and the electronic equipment that criminal investigation material evidence is saved from damage |
CN111442738A (en) * | 2020-03-23 | 2020-07-24 | 四川大学 | Device and method for acquiring three-dimensional trace characteristics of primer of cartridge case |
CN112085045B (en) * | 2020-04-07 | 2022-12-27 | 昆明理工大学 | Linear trace similarity matching algorithm based on improved longest common substring |
CN112085045A (en) * | 2020-04-07 | 2020-12-15 | 昆明理工大学 | Linear trace similarity matching algorithm based on improved longest common substring |
CN113624164A (en) * | 2021-08-06 | 2021-11-09 | 吉林省计量科学研究院 | Low-cost portable axle curved surface micro-scratch non-contact detection device and method |
CN113744238A (en) * | 2021-09-01 | 2021-12-03 | 南京工业大学 | Method for establishing bullet trace database |
CN113744238B (en) * | 2021-09-01 | 2023-08-01 | 南京工业大学 | Method for establishing bullet trace database |
CN116311628A (en) * | 2023-05-23 | 2023-06-23 | 合肥智辉空间科技有限责任公司 | Method and system for detecting safety performance of intelligent door lock |
CN116311628B (en) * | 2023-05-23 | 2023-08-11 | 合肥智辉空间科技有限责任公司 | Method and system for detecting safety performance of intelligent door lock |
Also Published As
Publication number | Publication date |
---|---|
CN109191502B (en) | 2021-08-24 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109191502A (en) | A kind of method of automatic identification shell case trace | |
CN109544612B (en) | Point cloud registration method based on feature point geometric surface description | |
CN104318548B (en) | Rapid image registration implementation method based on space sparsity and SIFT feature extraction | |
Jasiewicz et al. | Landscape similarity, retrieval, and machine mapping of physiographic units | |
CN107346550B (en) | It is a kind of for the three dimensional point cloud rapid registering method with colouring information | |
CN109409292A (en) | The heterologous image matching method extracted based on fining characteristic optimization | |
CN104680161A (en) | Digit recognition method for identification cards | |
CN105069790A (en) | Rapid imaging detection method for gear appearance defect | |
CN104880389B (en) | A kind of automatic measurement, sophisticated category method and its system of steel crystal grain mixed crystal degree | |
CN104778721A (en) | Distance measuring method of significant target in binocular image | |
CN107092871A (en) | Remote sensing image building detection method based on multiple dimensioned multiple features fusion | |
CN105787481B (en) | A kind of object detection method and its application based on the potential regional analysis of Objective | |
CN104636721A (en) | Palm print identification method based on contour and edge texture feature fusion | |
CN104680130A (en) | Chinese character recognition method for identification cards | |
Mostafa et al. | Shadow identification in high resolution satellite images in the presence of water regions | |
CN106815583A (en) | A kind of vehicle at night license plate locating method being combined based on MSER and SWT | |
CN107679470A (en) | A kind of traffic mark board detection and recognition methods based on HDR technologies | |
CN102446356A (en) | Parallel and adaptive matching method for acquiring remote sensing images with homogeneously-distributed matched points | |
CN108021890A (en) | A kind of high score remote sensing image harbour detection method based on PLSA and BOW | |
CN107644227A (en) | A kind of affine invariant descriptor of fusion various visual angles for commodity image search | |
CN103913166A (en) | Star extraction method based on energy distribution | |
Zaharieva et al. | Image based recognition of ancient coins | |
CN106372111A (en) | Local feature point screening method and system | |
CN105279522A (en) | Scene object real-time registering method based on SIFT | |
CN109584250A (en) | A kind of method that the visual zone of robust divides mark automatically |
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 |