CN106093946A

CN106093946A - A kind of target condensing method being applicable to scene surveillance radar and device

Info

Publication number: CN106093946A
Application number: CN201610359146.4A
Authority: CN
Inventors: 冯翔; 陈俊
Original assignee: Sichuan Jiuzhou ATC Technology Co Ltd
Current assignee: Sichuan Jiuzhou ATC Technology Co Ltd
Priority date: 2016-05-27
Filing date: 2016-05-27
Publication date: 2016-11-09
Anticipated expiration: 2036-05-27
Also published as: CN106093946B

Abstract

The present invention relates to radar data treatment technology, especially relate to a kind of target condensing method being applicable to scene surveillance radar and device.The present invention is directed to the problem that prior art exists, it is provided that a kind of target condensing method and device.Be set to system centre point in place with radar, radar power range is border, is a rectangular coordinate system plane by control space projection.Original point mark collection is assigned in corresponding space cell by coordinate figure the most in a direction, it is condensed space cell the most successively processing, obtain the cluster result in the direction, on this basis, primary cluster result is carried out the agglomeration process in another direction again, has just obtained final cohesion result.Only one-dimensional direction is carried out linear scanning owing to this algorithm once condenses, be better than the algorithm of traditional two dimensions of consideration, so referred to as dimension-reduction algorithm.Further, since the distance that need not calculate between any two points, its Algorithms T-cbmplexity is linear, and therefore efficiency is the highest.

Description

A kind of target condensing method being applicable to scene surveillance radar and device

Technical field

The present invention relates to radar data treatment technology, especially relate to a kind of target cohesion being applicable to scene surveillance radar Method and device.

Background technology

Scene surveillance radar is used for monitoring machine Ground traffic, to realize ground airborne vehicle and the tune of motor vehicles Degree, can send alarm in time when there is dangerous situation.The function of scene surveillance radar determines its high accuracy and real-time.Away from High Resolution reaches 3 meters, and azimuth resolution is less than or equal to 0.6 °, and the cycle is 1 second.During the targets such as radar scanning aircraft, automobile, record The target original point mark taken is usually present multiple, and the original point mark data of different target mix with false-alarm, and Targets Dots is coagulated Poly-process is exactly that original point mark data are sorted out, and the some mark data that same target is produced divide a set into and coagulate Poly-process produces unique Targets Dots, and rejects the some mark that clutter remnants, interference or inessential target cause, in order to after Continuous flight path processing.The flow chart of data processing figure of scene surveillance radar is as shown in Figure 1；

The essence of target cohesion is to sort out radar scanning video point, and the data belonging to same target divide into a collection Close.Conventional method has clustering algorithm, connected component labeling algorithm etc..

Cluster analysis finds data by limited Unlabeled data collection is divided into finite discrete " naturally " data set Structural information, is an important branch of unsupervised segmentation in statistical pattern classification.Cluster analysis is according to certain specific criteria One data set is divided into several different subclasses so that the sample similarity in same class is big as far as possible, inhomogeneity Sample variation the biggest.Scene surveillance radar system is high to the requirement of real-time.K-mean cluster, fuzzy poly- Class, although the Clustering Effect of these clustering algorithms is relatively good, but due to needs iteration, amount of calculation is the biggest.

Target coagulation problems can also be converted into the connected component labeling problem of bianry image.

Bianry image refers to only comprise the digital picture of background pixel and object pixel.For bianry image, the point of next-door neighbour I.e. it is considered connection.Bianry image connected component labeling refers to the target pixel points phase by meeting certain connection rule in image Same label shows.

Up to the present, conventional method for marking connected region mainly has following a few class:

1) the two-way method that is repeatedly scanned with: for the first time during scanning, each target pixel points is labeled as a unique label.Then, By forward and reverse scan labeled graph picture, and in the neighborhood of each pixel, propagate minimum label, until not having label to change Time till.

2) twice sweep method: for the first time during scanning, is stored in one and the equirotal two-dimemsional number of image by smporary label In group and formed of equal value right.During the end of scan, merge equivalent labels by certain searching method；During second time scanning, by equivalence Index value minimum in label gives the pixel that all equivalent labels are corresponding.Detailed arthmetic statement is: first to one two Value image is from left to right, be scanned from top to bottom.If the gray value of current pixel is 0, it is moved to next scanning position Put.If the gray value of current pixel is 1, check its left side, upper left, upper, these 4 neighbors of upper right.If above-mentioned 4 The gray value of individual neighbor is all 0, just to one new mark value of current pixel.If 4 neighbors only have one The gray value of individual pixel P is 1, just the mark value of P pixel is assigned to current pixel.If 4 neighbors have m The gray value of (1 < m≤4) individual pixel is 1, then according to a left side, upper left, upper, the priority of upper right, determine the labelling of current pixel Value.Then the mark value being had this m pixel is done of equal value right, and is classified to an equivalence in array.For the first time After the end of scan, all gray values be 1 point the most marked, but some labelling is of equal value.Arrange logarithm of equal value Group, equivalence to arranging as equivalence relation.Carrying out second time image scanning, the equivalence relation according to arranging gained is carried out again Labelling.For the second time after the end of scan, all gray values be 1 target area be all marked with different mark value.

3) region growth method: scan each pixel of bianry image successively.When finding certain unlabelled target picture During vegetarian refreshments, it is pressed into storehouse and starts repeatedly its region of labelling from this point, until storehouse is empty.Detailed arthmetic statement is: First, the bianry image of input is implemented progressive scan, finds first point in a unmarked region, this point of labelling.Checking should The eight neighborhood point of point labelling meet connectivity platform, and not yet labeled point, are recorded by newly-increased labelling point simultaneously It is used as the seed points of " region growth ".In follow-up labeling process, from the array of record seed points, constantly take out one Individual seed, implements aforesaid operations, so circulates, until the array of record seed points is empty.One connected component labeling terminates, and connects Again the next unmarked region of labelling, until all connected regions of input bianry image are the most labeled.

The operational efficiency of the most several conventional connected component labeling algorithms is the highest, and the two-way method that is repeatedly scanned with is due to scanning time Number too much, causes time efficiency the highest；Region growth method not only needs scanning repeatedly to object pixel, object pixel neighborhood of a point Number of comparisons is the most more, and needs the more stack manipulation time；Twice sweep method for step-like connected region due to Of equal value to too many and efficiency is the lowest.Secondly, these algorithms are strictly according to rank scanning, it is impossible to arrange ripple according to practical situation Door, motility is little.

Summary of the invention

The technical problem to be solved is: the problem existed for prior art, it is provided that one is applicable to scene The target condensing method of surveillance radar and device.Be set to system centre point in place with radar, radar power range is border, will Control space projection is a rectangular coordinate system plane.If then X-axis being divided into another relative direction from a direction Dry vertical banding spatial domain, as the elementary cell of X-direction cohesion；Again by Y-axis from a direction to another relative direction It is divided into some effect horizontal band-like spatial domains, as the elementary cell of Y-direction cohesion.During agglomeration process, first by be processed Original point trace set projects to system rectangular coordinate system plane, then carries out unidirectional agglomeration process successively.First by a side To coordinate figure original point mark collection is assigned in corresponding space cell, more successively space cell is processed, is somebody's turn to do Cluster result on direction, on this basis, carries out the agglomeration process (method in another direction again to primary cluster result Unanimously), just obtain final cohesion set, exported after the some set weighted average finally each condensed, obtain cluster knot Really.Only one-dimensional direction is carried out linear scanning owing to this algorithm once condenses, be better than the algorithm of traditional two dimensions of consideration, therefore And referred to as dimension-reduction algorithm.Further, since the distance that need not calculate between any two points, its Algorithms T-cbmplexity is linear, Therefore efficiency is the highest.

The technical solution used in the present invention is as follows:

A kind of target condensing method being applicable to scene surveillance radar includes:

Step 1: be set to system centre point in place with radar, radar power range (-MaxRange ~ MaxRange) is border, X Direction threshold value XThreshold is X-direction ultimate unit, and Y-direction threshold value YThreshold is direction elementary cell, will pipe Space projection processed is a rectangular coordinate system plane；Control space is drawn to another relative direction along X-axis from a direction It is divided into 2N vertically banding spatial domain band, as the elementary cell of X-direction cohesion, wherein N=MaxRange/Xthreshold；Again will Control space is divided into 2M horizontal band-like spatial domain band along Y-axis from a direction to another relative direction, coagulates as Y-direction Poly-elementary cell, wherein M=MaxRange/Ythreshold；

Step 2: obtain current time radar original point mark, carries out clutter process and obtains original point trace set to be condensed, perform step Rapid 3 or step 6 carry out unidirectional agglomeration process the most successively, the some mark after being condensed；

Step 3: original point trace set to be condensed is projected to rectangular coordinate system plane, treats cohesion original point trace set and carries out X Direction is condensed, it is thus achieved that X-direction congealing point trace set PointSet；Each congealing point trace set PointSet with index index is Key assignments, is stored in the cohesion original point trace set RawPointSets of X-direction；Index is X-direction spatial domain tape index belonging to point set Number；

Step 4: take a some trace set OnePointSet from cohesion original point trace set RawPointSets, it is carried out Y Direction agglomeration process, described Y-direction agglomeration process is identical with X-direction agglomeration process method；

After OnePointSet carries out the agglomeration process of Y-direction, the some trace set that cohesion produces all adds set LastPointSets；Judge that the some trace set condensing original point trace set RawPointSets is the most treated complete, if cohesion The point trace set of original point trace set RawPointSets has been disposed, and performs step 5, otherwise, from RawPointSets This trace set OnePointSet of middle deletion, performs step 4；

Step 5: the original point trace set in traversal LastPointSets, calculates the center of mass point of each set, i.e. weighted mean, Just the some mark after being condensed；

Step 6: original point trace set to be condensed is projected to rectangular coordinate system plane, treats cohesion original point trace set and carries out elder generation Carry out Y-direction cohesion, it is thus achieved that Y-direction congealing point trace set PointSet；Each congealing point trace set PointSet is with index Index is key assignments, is stored in the cohesion original point trace set RawPointSets of Y-direction；Index is that Y-direction belonging to point set is empty Territory tape index number；

Step 7: take a some trace set OnePointSet from cohesion original point trace set RawPointSets, it is carried out X Direction agglomeration process, described X-direction agglomeration process is identical with Y-direction agglomeration process method；

After OnePointSet carries out the agglomeration process of X-direction, the some trace set that cohesion produces all adds set LastPointSets；Judge that the some trace set condensing original point trace set RawPointSets is the most treated complete, if cohesion The point trace set of original point trace set RawPointSets has been disposed, and performs step 8, otherwise, from RawPointSets This trace set OnePointSet of middle deletion, performs step 7；

Step 8: the original point trace set in traversal LastPointSets, calculates the center of mass point of each set, i.e. weighted mean, Just the some mark after being condensed.

Further, described step 2 obtains current time radar original point mark, carry out clutter process obtain waiting condensing former Initial point trace set detailed process is:

Step 21: obtain current time radar original point mark as input；

Step 22: current time original point mark is carried out the cohesion of X-direction, it is thus achieved that X-direction merger point trace set PointSet, often Individual some trace set, with X-direction index index as key assignments, is stored in RawPointSets；

Step 23: obtain the original point trace set RawPointSets after X-direction cohesion；

Step 24: take a some trace set OnePointSet from RawPointSets, be ready for the agglomeration process of Y-direction； Described Y-direction agglomeration process is identical with X-direction agglomeration process method；

Step 25: OnePointSet carries out the agglomeration process of Y-direction, the some trace set that cohesion produces all adds set LastPointSets；

Step 26: judge that the some trace set of RawPointSets is the most treated complete, if RawPointSets is disposed, then Perform step 27, otherwise, perform step 24；

Step 27: obtain the set LastPointSets after merger；Original point trace set in traversal LastPointSets；If Point mark number in this trace set is more than threshold value iDensity, then added by the some mark in this trace set and treat congealing point mark Set；Otherwise, this trace set is deleted.

Further, treating cohesion original point mark data and carry out X-direction cohesion in described step 3, detailed process is:

Step 31: empty AllAarrayMap, obtain current time wait condense all original point mark points and be stored in data Caching, obtains unordered all original point mark points to be condensed；AllAarrayMap refers to reflecting of a trace set Penetrate relation；

Step 32: take the point of an original point mark to be condensed in points, calculates Descartes according to the polar coordinate of point and sits Mark, just will wait that with this condensing all original point mark points projects to rectangular coordinate system plane；X-coordinate according to point Point.x calculates spatial domain band belonging to point, index=point.x/XThreshold+1；

Step 33: point is stored in some trace set pointArray [index] that spatial domain band is corresponding, if current the most not this point Trace set, the most newly created one；With index as key assignments, pointArray [index] is value, and the mapping setting up some trace set is closed It is AllAarrayMap [index, pointArray [index]]；If last in point not all original point mark points One some mark, then delete point from points, perform step 32；

Step 34: take out the key assignments set keys of AllAarrayMap, index all in key assignments keys are carried out ascending order arrangement； Keys refers to the set of all index；

Step 35: travel through keys successively, takes next key value next of key assignments index and index；According to index, next Take out corresponding some trace set pointArray [index], some trace set pointArray [next] respectively；

Step 36: if pointArray [next] .Xmin-pointArray [index] .Xmax < Xthreshold, then perform Step 37, otherwise index=next, perform step 35；Wherein pointArray [index] .Xmax and pointArray [inext] .Xmin is pointArray [index] midpoint mark X-coordinate maximum and pointArray [next] midpoint mark X respectively Coordinate figure minima；

Step 37: the some mark in pointArray [next] is incorporated in pointArray [index], and updates PointArray [index] .Ymax=pointArray [next] .Ymax, deletes next corresponding in AllAarrayMap Storage record；Wherein pointArray [index] .Ymax is pointArray [index] midpoint mark Y coordinate maximum, PointArray [next] .Ymax is pointArray [next] midpoint mark Y-coordinate value maximum；

Step 38: if next arrives the maximum of keys, then perform step 310, otherwise perform step 39；

Step 39:next=next+1, jumps to step 36；

Step 310: in key value keys, index size carries out ascending order arrangement, is stored in the some trace set in AllAarrayMap RawPointSets。

Further, described step 4 is treated cohesion original point mark data and is carried out Y-direction cohesion, and detailed process is:

Step 41: empty AllAarrayMap, takes a some trace set OnePointSet from RawPointSets； AllAarrayMap refers to the mapping relations of a trace set；

Step 42: take the point of an original point mark to be condensed in OnePointSet, counts according to the Y coordinate point.y of point Calculate spatial domain band belonging to point, index=point. Y/Threshold+1；

Step 43: point is stored in some trace set pointArray [index] that spatial domain band is corresponding, if current the most not this point Trace set, the most newly created one；With index as key assignments, pointArray [index] is value, and the mapping setting up some trace set is closed It is AllAarrayMap [index, pointArray [index]]；If point is not last some mark of OnePointSet, Then delete point from OnePointSet, perform step 42；

Step 44: take out the key assignments set keys of AllAarrayMap, index all in key assignments keys are carried out ascending order arrangement； Keys refers to the set of all index；

Step 45: travel through keys successively, takes next key value next of key assignments index and index；According to index, next Take out corresponding some trace set pointArray [index], some trace set pointArray [next] respectively；

Step 46: if pointArray [next] .Ymin-pointArray [index] .Ymax < Ythreshold, then perform Step 47, otherwise, index=next, perform step 45；Wherein pointArray [index] .Ymax and pointArray [next] .Ymin is pointArray [index] midpoint mark Y coordinate maximum and pointArray [next] midpoint mark Y respectively Coordinate figure minima；

Step 47: the some mark in pointArray [next] is incorporated in pointArray [index], and updates PointArray [index] .Ymax=pointArray [next] .Ymax, deletes next corresponding in AllAarrayMap Storage record (wherein pointArray [index] .Ymax is pointArray [index] midpoint mark Y coordinate maximum, PointArray [next] .Ymax is pointArray [next] midpoint mark Y-coordinate value maximum)；

Step 48: if next arrives the maximum of keys, then perform step 410, otherwise perform step 49；

Step 49:next=next+1, jumps to step 46；

Step 410: in key value keys, index size carries out ascending order arrangement, is stored in the some trace set in AllAarrayMap LastPointSets；

Step 411: judge that OnePointSet has been last point of cohesion original point trace set RawPointSets Trace set, the most then the some trace set of cohesion original point trace set RawPointSets has been disposed, and performs step 5, no Then, from RawPointSets, delete OnePointSet, perform step 41.

Further, described step 5 travels through the original point trace set in LastPointSets, calculates barycenter, i.e. weight Meansigma methods, the point after being condensed, detailed process:

Step 51: obtain the original point trace set LastPointSets after cohesion；

Step 52: traversal LastPointSets, it is thus achieved that an original point trace set PointSet；

Step 53: calculating center of mass point pt of all original point marks in PointSet, the coordinate figure computing formula of pt is: the x-axis of pt Coordinate pt.x=∑ point.x/M, pt.y=∑ point.y/M, M are original point mark number in PointSet, the coordinate of barycenter pt It is the some mark coordinate after cohesion, pt is stored in output point mark caching MergedPointList；

Step 54: if LastPointSets has had stepped through, then terminate, MergedPointList is the some mark collection after cohesion Close, otherwise perform step 52.

A kind of target coacervation device being applicable to scene surveillance radar includes:

Rectangular coordinate system plane sets up module: be set to system centre point in place with radar, radar power range (-MaxRange ~ MaxRange) being border, X-direction threshold value XThreshold is X-direction ultimate unit, and Y-direction threshold value YThreshold is Y Direction elementary cell, is a rectangular coordinate system plane by control space projection；By control space along X-axis from a direction to phase To another direction be divided into 2N vertically banding spatial domain band, as the elementary cell of X-direction cohesion, wherein N= MaxRange/Xthreshold；Again control space is divided into 2M water from a direction to another relative direction along Y-axis Flat ribbon spatial domain band, as the elementary cell of Y-direction cohesion, wherein M=MaxRange/Ythreshold.

Filtering Processing and projection module: be used for obtaining current time radar original point mark, carry out clutter process and obtain waiting to coagulate Poly-original point trace set, is then carried out at unidirectional cohesion the most successively by X-Y cohesion module or Y-X cohesion resume module Reason, the some mark after being condensed；

X-Y condenses module: for original point trace set to be condensed is projected to rectangular coordinate system plane, treats cohesion original point mark Set carries out X-direction cohesion, it is thus achieved that X-direction congealing point trace set PointSet；Each congealing point trace set PointSet is with rope Drawing index is key assignments, is stored in the cohesion original point trace set RawPointSets of X-direction；Index is X-direction belonging to point set Spatial domain tape index number；From cohesion original point trace set RawPointSets take a some trace set OnePointSet, prepare into Row Y-direction agglomeration process；Described Y-direction agglomeration process is identical with X-direction agglomeration process method；

Then OnePointSet carries out the agglomeration process of Y-direction, and the some trace set that cohesion produces all adds set LastPointSets；Judge that the some trace set condensing original point trace set RawPointSets is the most treated complete, if cohesion The point trace set of original point trace set RawPointSets has been disposed, then the original point in traversal LastPointSets Trace set, calculates barycenter, i.e. weighted mean, the original point after being condensed；Otherwise, from RawPointSets, take a point Trace set OnePointSet, is ready for the agglomeration process of Y-direction.

Y-X condenses module: for original point trace set to be condensed is projected to rectangular coordinate system plane, treats cohesion original Point trace set carries out Y-direction cohesion, it is thus achieved that Y-direction congealing point trace set PointSet；Each congealing point trace set PointSet With index index as key assignments, it is stored in the cohesion original point trace set RawPointSets of Y-direction；Index is Y belonging to point set Spatial domain, direction tape index number；A point trace set OnePointSet is taken from cohesion original point trace set RawPointSets, accurate For carrying out X-direction agglomeration process；Described X-direction agglomeration process is identical with Y-direction agglomeration process method；The most right OnePointSet carries out the agglomeration process of X-direction, and the some trace set that cohesion produces all adds set LastPointSets；Sentence The point trace set of disconnected cohesion original point trace set RawPointSets is the most treated complete, if cohesion original point trace set The point trace set of RawPointSets has been disposed, then the original point trace set in traversal LastPointSets, calculates matter The heart, i.e. weighted mean, the original point after being condensed；Otherwise, from RawPointSets, take a some trace set OnePointSet, is ready for the agglomeration process of Y-direction.

Further, described Filtering Processing and projection module obtain current period radar original point mark, carry out clutter process Obtain original point trace set to be condensed, specifically include:

Step 71: obtain current time radar original point mark as input；

Step 72: current time original point mark is carried out the cohesion of Y-direction, it is thus achieved that Y-direction merger point trace set PointSet, often Individual some trace set, with Y-direction index index as key assignments, is stored in RawPointSets；

Step 73: obtain the original point trace set RawPointSets after Y-direction cohesion；

Step 74: take a some trace set OnePointSet from RawPointSets, be ready for the agglomeration process of Y-direction； Described X-direction agglomeration process is identical with Y-direction agglomeration process method；

Step 75: OnePointSet carries out the agglomeration process of X-direction, the some trace set that cohesion produces all adds set LastPointSets；

Step 76: judge that the some trace set of RawPointSets is the most treated complete, if RawPointSets is disposed, then Perform step 77, otherwise, perform step 74；

Step 77: obtain the set LastPointSets after merger；Original point trace set in traversal LastPointSets；If Point mark number in this trace set is more than threshold value iDensity, then added by the some mark in this trace set and treat congealing point mark Set；Otherwise, this trace set is deleted.

Further, described step X-Y cohesion module is treated cohesion original point mark data and carry out Y-direction cohesion, specifically Process is:

Step 81: empty AllAarrayMap, obtain current time wait condense all original point mark points and be stored in data Caching, obtains unordered all original point mark points to be condensed；AllAarrayMap refers to reflecting of a trace set Penetrate relation；

Step 82: take the point of an original point mark to be condensed in points, calculates Descartes according to the polar coordinate of point and sits Mark, just will wait that with this condensing all original point mark points projects to rectangular coordinate system plane；Y coordinate according to point Point.y calculates spatial domain band belonging to point, index=point.y/YThreshold+1；

Step 83: point is stored in some trace set pointArray [index] that spatial domain band is corresponding, if current the most not this point Trace set, the most newly created one；With index as key assignments, pointArray [index] is value, and the mapping setting up some trace set is closed It is AllAarrayMap [index, pointArray [index]]；If last in point not all original point mark points One some mark, then delete point from points, perform step 82；

Step 84: take out the key assignments set keys of AllAarrayMap, index all in key assignments keys are carried out ascending order arrangement； Keys refers to the set of all index；

Step 85: travel through keys successively, takes next key value next of key assignments index and index；According to index, next Take out corresponding some trace set pointArray [index], some trace set pointArray [next] respectively；

Step 86: if pointArray [next] .Ymin-pointArray [index] .Ymax < Ythreshold, then perform Step 87, otherwise index=next, perform step 85；Wherein pointArray [index] .Ymax and pointArray [inext] .Ymin is pointArray [index] midpoint mark Y coordinate maximum and pointArray [next] midpoint mark Y respectively Coordinate figure minima；

Step 87: the some mark in pointArray [next] is incorporated in pointArray [index], and updates PointArray [index] .Ymax=pointArray [next] .Ymax, deletes next corresponding in AllAarrayMap Storage record (wherein pointArray [index] .Ymax is pointArray [index] midpoint mark Y coordinate maximum, PointArray [next] .Ymax is pointArray [next] midpoint mark Y-coordinate value maximum)；

Step 88: if next arrives the maximum of keys, then perform step 810, otherwise perform step 89；

Step 89:next=next+1, jumps to step 36；

Step 810: in key value keys, index size carries out ascending order arrangement, is stored in the some trace set in AllAarrayMap RawPointSets。

Further, described X-Y cohesion module is treated cohesion original point mark data and is carried out X-direction cohesion, and detailed process is:

Step 91: empty AllAarrayMap, takes a some trace set OnePointSet from RawPointSets； AllAarrayMap refers to the mapping relations of a trace set；

Step 92: take the point of an original point mark to be condensed in OnePointSet, counts according to X-coordinate point.x of point Calculate spatial domain band belonging to point, index=point.x/XThreshold+1；

Step 93: point is stored in some trace set pointArray [index] that spatial domain band is corresponding, if current the most not this point Trace set, the most newly created one；With index as key assignments, pointArray [index] is value, and the mapping setting up some trace set is closed It is AllAarrayMap [index, pointArray [index]]；If point is not last point in OnePointSet Mark, then delete point from OnePointSet, perform step 92；

Step 94: take out the key assignments set keys of AllAarrayMap, index all in key assignments keys are carried out ascending order arrangement； Keys refers to the set of all index；

Step 95: travel through keys successively, takes next key value next of key assignments index and index；According to index, next Take out corresponding some trace set pointArray [index], some trace set pointArray [next] respectively；

Step 96: if pointArray [next] .Xmin-pointArray [index] .Xmax < Xthreshold, then perform Step 97, otherwise, index=next, perform step 95；Wherein pointArray [index] .Xmax and pointArray [next] .Xmin is pointArray [index] midpoint mark X-coordinate maximum and pointArray [next] midpoint mark X respectively Coordinate figure minima；

Step 97: the some mark in pointArray [next] is incorporated in pointArray [index], and updates PointArray [index] .Xmax=pointArray [next] .Xmax, deletes next corresponding in AllAarrayMap Storage record (wherein pointArray [index] .Xmax is pointArray [index] midpoint mark X-coordinate maximum, PointArray [next] .Xmax is pointArray [next] midpoint mark X-coordinate value maximum)；

Step 98: if next arrives the maximum of keys, then perform step 910, otherwise perform step 99；

Step 99:next=next+1, jumps to step 96；

Step 910: in key value keys, index size carries out ascending order arrangement, is stored in the some trace set in AllAarrayMap LastPointSets；

Step 911: judge that OnePointSet has been last point of cohesion original point trace set RawPointSets Trace set, the most then the some trace set of cohesion original point trace set RawPointSets has been disposed, traversal Original point trace set in LastPointSets, calculating barycenter, i.e. weighted mean, the point after being condensed, otherwise, from RawPointSets deletes this trace set OnePointSet, performs step 91.

Further, described step X-YX-Y cohesion module travels through the original point trace set in LastPointSets, meter Calculation barycenter, i.e. weighted mean, the point after being condensed, detailed process:

Step 101: obtain the original point trace set LastPointSets after cohesion；

Step 102: traversal LastPointSets, it is thus achieved that an original point trace set PointSet；

Step 103: calculating center of mass point pt of all original point marks in PointSet, the coordinate figure computing formula of pt is: the x-axis of pt Coordinate pt.x=∑ point.x/M, pt.y=∑ point.y/M, M are original point mark number in PointSet, the coordinate of barycenter pt It is the some mark coordinate after cohesion, pt is stored in output point mark caching MergedPointList；

Step 104: if LastPointSets has had stepped through, then terminate, MergedPointList is the some mark after cohesion Set, otherwise performs step 102.

In sum, owing to have employed technique scheme, the invention has the beneficial effects as follows:

Be set to system centre point in place with radar, radar power range is border, is that a right angle is sat by control space projection Mark system plane.Then X-axis is divided into several vertical banding spatial domains, as X side to another relative direction from a direction Elementary cell to cohesion；Again Y-axis is divided into several horizontal band-like spatial domains from a direction to another relative direction, makees Elementary cell for Y-direction cohesion.During agglomeration process, first original point trace set to be processed is projected to system rectangular coordinate It is plane, then carries out unidirectional agglomeration process successively.Original point mark collection is assigned to phase by coordinate figure the most in a direction In the space cell answered, more successively space cell is processed, obtain the cluster result in the direction, on this basis, right Primary cluster result carries out the agglomeration process (method is consistent) in another direction again, has just obtained final cohesion set, Export after the some set weighted average finally each condensed, obtain cluster result.Owing to this algorithm once condenses only to one-dimensional Direction carries out linear scanning, is better than the algorithm of traditional two dimensions of consideration, owing to this algorithm once condenses only to one-dimensional direction Carry out linear scanning, be better than the algorithm of traditional two dimensions of consideration, so referred to as dimension-reduction algorithm.Further, since need not calculate Distance between any two points, its Algorithms T-cbmplexity is linear, and therefore efficiency is the highest.

Accompanying drawing explanation

Examples of the present invention will be described by way of reference to the accompanying drawings, wherein:

Fig. 1 is the flow chart of data processing figure of scene surveillance radar in prior art.

Fig. 2 is dimensionality reduction clustering algorithm schematic diagram of the present invention.

Fig. 3 is inventive algorithm flow chart.

Fig. 4 is the flow chart that clutter is filtered by embodiment illustrated in fig. 3 step 302.

Fig. 5 is the X-direction cohesion schematic flow sheet of embodiment illustrated in fig. 3 step 303.

Fig. 6 a is that X-axis spatial domain band divides schematic diagram.

Fig. 6 b is that Y-axis spatial domain band divides schematic diagram.

Fig. 7 is the schematic diagram of embodiment illustrated in fig. 3 step 309.

Detailed description of the invention

All features disclosed in this specification, or disclosed all methods or during step, except mutually exclusive Feature and/or step beyond, all can combine by any way.

Any feature disclosed in this specification, unless specifically stated otherwise, all can by other equivalence or there is similar purpose Alternative features is replaced.I.e., unless specifically stated otherwise, an example during each feature is a series of equivalence or similar characteristics ?.

Scene surveillance radar general work is in scanning limit, limit tracking mode, and the main object of observation is the motion mesh on airport Mark, such as aircraft and automobile, the radar return of this kind of target is detected and is formed original point mark, obtain be not one but One group of some mark data, Target dots processing seeks to filter original point mark data, merger and cohesion, obtains and characterizes target physical The Plot coherence estimated value of location parameter a, it is desirable to target only produces an Accurate Points mark.It is further that some mark data process Requirement is: suppression False Intersection Points mark, it is provided that the isoparametric estimation accuracy in target range, orientation, and can effectively Targets Dots with Clutter point mark distinguishes, and creates a good environment for follow-up Track In Track.

Algorithm arrangement is as follows: be set to system centre point in place with radar, and radar power range is border, by control space It is projected as a rectangular coordinate system plane.Then it is divided into several vertical from a direction to another relative direction X-axis Banding spatial domain, as the elementary cell of X-direction cohesion；Again Y-axis is divided into some from a direction to another relative direction Individual horizontal band-like spatial domain, as the elementary cell of Y-direction cohesion.During agglomeration process, first by original point trace set to be processed Project to system rectangular coordinate system plane, then carry out unidirectional agglomeration process successively.Coordinate figure the most in a direction will Original point mark collection is assigned in corresponding space cell, then processes space cell successively, obtains the cluster in the direction As a result, on this basis, primary cluster result is carried out the agglomeration process (method is consistent) in another direction again, just obtains Final cohesion set, exports after the some set weighted average finally each condensed, obtains cluster result.Due to this algorithm Once cohesion only carries out linear scanning to one-dimensional direction, is better than the algorithm of traditional two dimensions of consideration, so referred to as dimensionality reduction is calculated Method.Further, since the distance that need not calculate between any two points, its Algorithms T-cbmplexity is linear, and therefore efficiency is the highest. Flow chart is as shown in Figure 2.

Step one, being set to system centre point in place with radar, radar power range is border, by control space projection is One rectangular coordinate system plane；

Step 2, X-axis is divided into several vertical banding spatial domains, as X-direction to another relative direction from a direction The elementary cell of cohesion；

Step 3, again Y-axis is divided into several horizontal band-like spatial domains from a direction to another relative direction, as Y side Elementary cell to cohesion；

Step 4, acquisition current period radar original point mark data, project to system right angle by original point trace set to be processed Coordinate plane, then carry out unidirectional agglomeration process successively, can first X-direction agglomeration process Y-direction agglomeration process again, or first Y-direction agglomeration process X-direction agglomeration process again, below as a example by the agglomeration process of first X-direction Y-direction again；

Step 5, laterally sequence: the ripple door WaveThreshold _ X in x direction is set, with ripple door for standard size by thunder Reaching scan data and map storage by x coordinate subdivision, sequence number is corresponding with storage list；Scan data is divided into M by x coordinate Individual junior unit, records minima and the maximum of x coordinate in each junior unit simultaneously；

Step 6, laterally cohesion: on x direction, compare X (i) max and X (i+1) the min value in adjacent cells, If difference is less than thresholding WaveThreshold _ X, then later unit being incorporated to previous element, after deleting, one is single simultaneously Unit, completes laterally to condense；

Longitudinally sequence in step 7, unit: the ripple door size WaveThreshold _ Y in y direction is set, each to transversely Unit X1, X2 ..., XM carries out longitudinal sequence respectively in unit.Such as, after carrying out X1 longitudinally sorting, sit according to y Unit is divided into N set 1Y1,1Y2 by mark ..., 1YN, the rule of pressing in unit is mapped to these small sets.With Time record minima and the maximum of y coordinate in each small set；

Longitudinally cohesion in step 8, unit: on y direction, compare Y (j) max and Y (j+1) min of adjacent cells Unit two, if difference is less than WaveThreshold _ Y, are merged by value, will integrate with previous set by a collection afterwards, delete simultaneously Except a rear set, after longitudinally condensing in completing certain unit, the point coordinates in gathering is weighted averagely, statistics point number, If greater than iDensity, then export weighted mean, otherwise delete.The point of output carries out follow-up flight path processing；

Step 9, method according to step 8, be sequentially completed X1, X 2 ..., longitudinally cohesion in the unit of XM, finally when The cohesion of front cycle radar original point mark data completes.

Only one-dimensional direction is carried out linear scanning owing to this algorithm once condenses, be better than the calculation of traditional two dimensions of consideration Method, so referred to as dimension-reduction algorithm.Reduce computational complexity by dimensionality reduction, two dimension computing is converted into one-dimensional operation, secondly, no By the distance calculated between any two points, therefore its Algorithms T-cbmplexity is linear, and efficiency is the highest.

In order to debug conveniently, can be by setting the parameters to the accuracy and speed of adjustment algorithm, three most important parameters are: WaveThreshold _ X, WaveThreshold _ Y and iDensity.WaveThreshold _ X is laterally sequence and coagulates Ripple door size time poly-, WaveThreshold _ Y be in unit longitudinally sequence and cohesion time ripple door size, iDensity It it is the density thresholding of output point.IDensity is had a mark by 0 output, and iDensity may be considered and filters clutter point One thresholding.Emulation shows, under identical running environment, uses conventional connected component labeling algorithm, can only locate each second Manage more than 500 target (including the subsequent treatment such as flight path processing), use the target disposal ability of dimensionality reduction clustering algorithm up to number 100000 points/second.

Specific embodiment:

Fig. 3 for the present invention provide a kind of divide based on two-way spatial domain band carry out fast target cohesion dimensionality reduction clustering algorithm Embodiment schematic flow sheet, comprises the following steps:

Step 301: obtain the original point mark that current time is to be condensed；

Step 302: clutter filters, rejects the original point mark that clutter remnants, interference or inessential target cause；

Step 303: treat cohesion original point mark and carry out the cohesion of X-direction, it is thus achieved that X-direction each congealing point trace set, each some mark Set PointSet is with spatial domain reel number belonging to index index(point set) as key assignments, it is stored in RawPointSets；

Step 304: obtain the original point trace set RawPointSets after X-direction cohesion；

Step 305: take a some trace set OnePointSet from RawPointSets, be ready at the cohesion of Y-direction Reason；

Step 306: OnePointSet carries out the agglomeration process of Y-direction, the some trace set that cohesion produces all adds set LastPointSets；

Step 307: judge that the some trace set of RawPointSets is the most treated complete, the most then execution step 308, otherwise, Perform step 305；

Step 308: the original point trace set in traversal LastPointSets, calculates barycenter, namely weighted mean, just obtains Arrive the point after cohesion.

Step 309: the point after output cohesion.

Fig. 4 is the flow chart that clutter is filtered by embodiment illustrated in fig. 3 step 302.For rejecting clutter remnants, interference or nothing Close the some mark that critical target causes, in order to avoid affecting subsequent treatment.Algorithm and Plot coherence that clutter filters are basically identical (in detail Algorithm steps sees below literary composition), need first the original point mark of current time to be carried out merger, be divided into original point trace set one by one, If the some mark number that difference is in certain original point trace set is less than threshold value, it is judged to clutter, the original point in this set Mark is deleted entirely.Clutter processes and can be arranged as required to the parameter that clutter filtration needs.

RawPointSets, OnePointSet, LastPointSets are the temporary variable in algorithm, algorithm include with Lower step:

Step 401: current time original point trace set to be condensed is as input；

Step 402: treat cohesion original point trace set and carry out the merger of X-direction, it is thus achieved that X-direction merger point trace set PointSet, each some trace set is with spatial domain reel number belonging to index index(point set) as key assignments, it is stored in RawPointSets In；

Step 403: obtain the original point trace set RawPointSets after X-direction merger；

Step 404: take a some trace set OnePointSet from RawPointSets, be ready at the merger of Y-direction Reason；

Step 405: the merger that OnePointSet carries out Y-direction processes, method is with step 402, the some trace set that merger produces All add set LastPointSets；

Step 406: judge that RawPointSets is the most treated complete, the most then perform step 407, otherwise, perform step 404；；

Step 407: obtain the set LastPointSets after merger；

Step 408: the original point trace set in traversal LastPointSets；

Step 409: if the some mark number in this trace set is more than threshold value iDensity, then perform step 410, otherwise, hold Row step 411；

Step 410: the some mark in this trace set is added and treats congealing point trace set；

Step 411: delete this trace set.

Fig. 5 is the schematic flow sheet of embodiment illustrated in fig. 3 step 303, treats cohesion original point trace set and carries out X-direction Cohesion, Y-direction agglomeration process flow process is identical.Comprise the following steps:

Step 501: divide spatial domain band, is set to center in place with radar, actual radar coverage (-MaxRange ~ MaxRange) being limited, X-direction threshold value XThreshold is ultimate unit, and along X-axis, control space plane is divided into some bands Shape region, number of regions is 2N, N=MaxRange/XThreshold, and Y-direction spatial domain band divides similar, and number of regions is 2M, M= MaxRange/YThreshold, as shown in Figure 6;

Step 502: obtain the original point mark that current time is to be condensed, be stored in data buffer storage；

Step 503: coordinate transform, takes an original point mark point, calculates cartesian coordinate according to its polar coordinate；

Step 504: calculating spatial domain band belonging to point according to the X-coordinate of point, spatial domain tape index number is index=point.x/ XThreshold+1；

Step 505: point is stored in some trace set pointArray [index] that spatial domain band is corresponding, if current the most not this point Trace set, the most newly created one；

Step 506: with index as key, pointArray [index] is value, sets up the mapping relations of some trace set AllAarrayMap [index, pointArray [index]], calculates and is had the minimum of a mark, maximum X to sit in pointArray Mark xmin, xmax；

Step 507: take out the key assignments set keys of AllAarrayMap, in ascending order；

Step 508: travel through keys successively, takes a key assignments key and its next key assignments next, takes out they correspondences respectively Point trace set X [key], X [next]；

Step 509: judge whether X [next] .xmin-X [key] .xmax is less than threshold value Xthreshold, if it is not, perform Step 510, if so, performs step 511；

Step 510:key=next, jumps to step 508 and starts to perform；

Step 511: the some mark in X [next] is incorporated in X [key], and updates X [key] .xmax=X [next] .xmax, AllAarrayMap deletes storage record corresponding to next；

Step 512: if next arrives the end of keys, then perform step 514；

Step 513:next=next+1, jumps to step 509 and performs；

Step 514: key value ascending order, is stored in RawPointSets by the some trace set in AllAarrayMap.

Fig. 7 is the schematic flow sheet of embodiment illustrated in fig. 3 step 308, and the original point mark that will be deemed as same target merges It is a some mark, comprises the following steps:

Step 701: obtain the original point trace set LastPointSets after cohesion；

Step 702: traversal LastPointSets, it is thus achieved that an original point trace set PointSet；

Step 703: calculating center of mass point pt of original point mark in PointSet, the coordinate figure computing formula of pt is: pt.x=∑ Point.x/N, pt.y=∑ point.y/N, N are original point mark number in PointSet, and the coordinate of pt is the some mark after cohesion Coordinate, is stored in output point mark caching MergedPointList by pt；

Step 704: if LastPointSets has had stepped through, then terminate, otherwise performs step 702；

The invention is not limited in aforesaid detailed description of the invention.The present invention expands to any new spy disclosed in this manual Levy or any new combination, and the arbitrary new method that discloses or the step of process or any new combination.

Claims

1. the target condensing method being applicable to scene surveillance radar, it is characterised in that including:

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 1, it is characterised in that described Step 2 obtains current time radar original point mark, carries out clutter and process and obtain original point trace set detailed process to be condensed and be:

Step 21: obtain current time radar original point mark as input；

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 1, it is characterised in that described Treating cohesion original point mark data in step 3 and carry out X-direction cohesion, detailed process is:

Step 31: empty AllAarrayMap, obtain current time wait condense all original point mark points and be stored in data Caching, obtains unordered all original point mark points to be condensed；AllAarrayMap refers to reflecting of a trace set Penetrating relation, with spatial domain reel number index belonging to a mark as key, the some trace set being positioned at this spatial domain band is value；

Step 39:next=next+1, jumps to step 36；

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 1, it is characterised in that described Step 4 is treated cohesion original point mark data and is carried out Y-direction cohesion, and detailed process is:

Step 41: empty AllAarrayMap, takes a some trace set OnePointSet from RawPointSets； AllAarrayMap refers to the mapping relations of a trace set, as key, is positioned at this sky with spatial domain reel number index belonging to a mark The point trace set of territory band is value；

Step 42: take the point of an original point mark to be condensed in OnePointSet, counts according to the Y coordinate point.y of point Calculate spatial domain band belonging to point, index=point.y/YThreshold+1；

Step 49:next=next+1, jumps to step 46；

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 1, it is characterised in that described Step 5 travels through the original point trace set in LastPointSets, calculates barycenter, i.e. weighted mean, after being condensed Point, detailed process:

Step 51: obtain the original point trace set LastPointSets after cohesion；

6. the target coacervation device being applicable to scene surveillance radar, it is characterised in that including:

Rectangular coordinate system plane sets up module: be set to system centre point in place with radar, radar power range (-MaxRange ~ MaxRange) being border, X-direction threshold value XThreshold is X-direction ultimate unit, and Y-direction threshold value YThreshold is Y Direction elementary cell, is a rectangular coordinate system plane by control space projection；By control space along X-axis from a direction to phase To another direction be divided into 2N vertically banding spatial domain band, as the elementary cell of X-direction cohesion, wherein N= MaxRange/Xthreshold；Again control space is divided into 2M water from a direction to another relative direction along Y-axis Flat ribbon spatial domain band, as the elementary cell of Y-direction cohesion, wherein M=MaxRange/Ythreshold；

Filtering Processing and projection module: be used for obtaining current time radar original point mark, carry out clutter process obtain waiting condensing former Initial point trace set, then carries out unidirectional agglomeration process the most successively by X-Y cohesion module or Y-X cohesion resume module, Point mark after cohesion；

Then OnePointSet carries out the agglomeration process of Y-direction, and the some trace set that cohesion produces all adds set LastPointSets；Judge that the some trace set condensing original point trace set RawPointSets is the most treated complete, if cohesion The point trace set of original point trace set RawPointSets has been disposed, then the original point in traversal LastPointSets Trace set, calculates barycenter, i.e. weighted mean, the original point after being condensed；Otherwise, from RawPointSets, take a point Trace set OnePointSet, is ready for the agglomeration process of Y-direction；

Y-X condenses module: for original point trace set to be condensed is projected to rectangular coordinate system plane, treats cohesion original point mark Set carries out Y-direction cohesion, it is thus achieved that Y-direction congealing point trace set PointSet；Each congealing point trace set PointSet is with rope Drawing index is key assignments, is stored in the cohesion original point trace set RawPointSets of Y-direction；Index is Y-direction belonging to point set Spatial domain tape index number；From cohesion original point trace set RawPointSets take a some trace set OnePointSet, prepare into Row X-direction agglomeration process；Described X-direction agglomeration process is identical with Y-direction agglomeration process method；Then OnePointSet is entered The agglomeration process of row X-direction, the some trace set that cohesion produces all adds set LastPointSets；Judge cohesion original point The point trace set of trace set RawPointSets is the most treated complete, if the point of cohesion original point trace set RawPointSets Trace set has been disposed, then the original point trace set in traversal LastPointSets, calculates barycenter, i.e. weighted mean, Original point after being condensed；Otherwise, from RawPointSets, take a some trace set OnePointSet, be ready for Y side To agglomeration process.

A kind of target coacervation device being applicable to scene surveillance radar the most according to claim 6, it is characterised in that described Filtering Processing and projection module obtain current period radar original point mark, carry out clutter process and obtain original point mark collection to be condensed Close, specifically include:

Step 71: obtain current time radar original point mark as input；

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 6, it is characterised in that described Treating cohesion original point mark data in step X-Y cohesion module and carry out Y-direction cohesion, detailed process is:

Step 81: empty AllAarrayMap, obtain current time wait condense all original point mark points and be stored in data Caching, obtains unordered all original point mark points to be condensed；AllAarrayMap refers to reflecting of a trace set Penetrating relation, with spatial domain reel number index belonging to a mark as key, the some trace set being positioned at this spatial domain band is value；

Step 89:next=next+1, jumps to step 36；

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 6, it is characterised in that described X-Y cohesion module is treated cohesion original point mark data and is carried out X-direction cohesion, and detailed process is:

Step 91: empty AllAarrayMap, takes a some trace set OnePointSet from RawPointSets； AllAarrayMap refers to the mapping relations of a trace set, as key, is positioned at this sky with spatial domain reel number index belonging to a mark The point trace set of territory band is value；

Step 99:next=next+1, jumps to step 96；

A kind of target condensing method being applicable to scene surveillance radar the most according to claim 6, it is characterised in that described Step X-YX-Y cohesion module travels through the original point trace set in LastPointSets, calculates barycenter, i.e. weighted mean, Point after being condensed, detailed process:

Step 101: obtain the original point trace set LastPointSets after cohesion；