CN109815784A - A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium - Google Patents
A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium Download PDFInfo
- Publication number
- CN109815784A CN109815784A CN201811444291.8A CN201811444291A CN109815784A CN 109815784 A CN109815784 A CN 109815784A CN 201811444291 A CN201811444291 A CN 201811444291A CN 109815784 A CN109815784 A CN 109815784A
- Authority
- CN
- China
- Prior art keywords
- polarization
- thermal infrared
- data
- infrared imager
- pixel
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Landscapes
- Image Processing (AREA)
- Closed-Circuit Television Systems (AREA)
- Radiation Pyrometers (AREA)
Abstract
The invention discloses a kind of intelligent method for classifying based on thermal infrared imager, it include: two polarization process steps, obtain infrared original uncorrected data, each object contour pixel point set is counted, two polarization datas of corresponding dynamic object object and two polarization datas of corresponding static object object are obtained by mapping and two polarization operations；Template library compares step, by two polarization datas of corresponding dynamic object object, compares dynamic instrumentation template library, judges targets of type；By two polarization datas of corresponding static object object, static instrumentation template library is compared, judges targets of type.The invention also discloses a kind of intelligent classification system and storage medium based on thermal infrared imager, a kind of intelligent method for classifying based on thermal infrared imager, system and storage medium of the invention can be realized the intelligent recognition and classification of object.It is not influenced simultaneously by complex environments such as light luminance, shades, meets the object detection classification of varying environment.
Description
Technical field
The present invention relates to monitoring detection techniques field more particularly to a kind of intelligent method for classifying based on thermal infrared imager,
System and storage medium.
Background technique
Target identification and classification are a basic project of field of image processing and field of machine vision, people's warp in life
Often some objects are identified using eyes and brain in study, Working Life.With the twentieth century computer forties invention and
Deepen continuously research of the mankind to artificial intelligence technology, the highly desirable role for playing the part of eyes by imaging sensor of people borrow
Help computer to replace or assist the mankind to carry out mental labour.Therefore, research institution both domestic and external or scholar are in target identification
Many researchs have been done in this field, and target identification technology is grown rapidly, and has many research achievements to be applied in practice.In recent years
Come, with the further investigation to image processing techniques and mode identification technology, so that the method using machine vision identifies ground
Target (vehicle, pedestrian or recognition of face) becomes possibility.
Currently, traditional monitoring device is all based on visible light sensor, these equipment are in illumination, the good item of weather
Clearly picture can be collected under part, but in the case where night, rain and snow weather and dense smoke thick fog block, they are difficult to
Obtain effective target information.And it is directed to the intelligent recognition classification method of object, it common are point based on template matching
Class method, the classification method based on model, the classification method based on artificial neural network and be based on nuclear machine learning method, this
Class target intelligent recognition algorithm, the method for being all based on machine learning training template library, extracts and counts the number of clarification of objective
According to come the identification of realizing target, but the above method is realized with radar or high-definition camera etc. mostly, is merely made
With these recognition methods, the computing capability of hardware is required high, and is easy to be influenced by light and environment, real-time compared with
Difference.However, often the application under these extreme conditions is only the emphasis for needing to pay close attention to, single visible light sensor is only used
Traditional method for identifying and classifying is cooperated to be no longer satisfied the demand intelligently detected.And infrared imagery technique relies on its anti-interference
Good, passive work and the strong feature of target identification ability successfully compensate for visible light sensor under the conditions of exceedingly odious not
The shortcomings that capable of being imaged very well, so that infrared image processing technology becomes the key technology in current information processing, and is examined in pedestrian
It is fully used in the fields such as survey, vehicle identification and aerial survey.
Based on this, a kind of intelligent Target identification and classification method based on thermal infrared imager is provided, Lai Shixian object
Intelligent recognition and classification, and the influence of the complex environments such as light luminance, shade can be excluded well, it is current intelligent monitoring
The technical issues of detection techniques field value must probe into.
Summary of the invention
For overcome the deficiencies in the prior art, one of the objects of the present invention is to provide a kind of intelligence based on thermal infrared imager
Energy classification method, can be realized the intelligent recognition and classification of object.
An object of the present invention adopts the following technical scheme that realization:
A kind of intelligent method for classifying based on thermal infrared imager, comprising: two polarization process steps obtain infrared original naked number
According to counting each object contour pixel point set, obtain two poles of corresponding dynamic object object by mapping and two polarization operations
Change two polarization datas of data and corresponding static object object；Template library compares step, by two polarization numbers of corresponding dynamic object object
According to comparison dynamic instrumentation template library judges targets of type；By two polarization datas of corresponding static object object, static visit is compared
Template library is surveyed, judges targets of type.
Further, it in the two polarization process step, to the pixel collection of dynamic object object, is drawn after piecemeal processing
Dynamic object object contour pixel point set is separated, object tile data is intercepted out according to rectangular edges pixel position, into
Row mapping and two polarization operations.
Further, in the two polarization process step, object tile data is mapped in pixel matrix
Carry out two polarization operations.
Further, in the two polarization process step, two polarization operations of corresponding dynamic object object include object
Two polarization and two polarization of object high/low temperature of shape.
Further, in the two polarization process step, for static object object, by infrared original uncorrected data figure
The contour pixel point set of corresponding static object object is obtained as carrying out edge extracting, by carrying out two polarization operations after prescreening.
Further, in the two polarization process step, two polarization operations of static object object are the polarization of shape two.
Further, it is compared in step in the template library, if two polarization datas of dynamic object object mismatch dynamic and visit
Targets of type any in template library is surveyed, then triggers abnormal object alarm.
Further, it is compared in step in the template library, when judging targets of type, while recording the position of object
It sets, is superimposed by OSD, real-time mark is carried out to target on infrared video stream, and count on after the quantity of corresponding object
Reach client.
The second object of the present invention is to provide a kind of intelligent classification system based on thermal infrared imager, can be realized target
The intelligent recognition and classification of object.
The second object of the present invention adopts the following technical scheme that realization:
A kind of intelligent classification system based on thermal infrared imager, including infrared detector and FPGA module, the infrared spy
Device is surveyed to be used to obtain the infrared original uncorrected data investigated in field angle, the FPGA module and the infrared detector communication link
It connects, realizes a kind of intelligent method for classifying based on thermal infrared imager as described in one of the object of the invention.
The third object of the present invention is to provide a kind of storage medium, can be realized the intelligent recognition and classification of object.
The third object of the present invention adopts the following technical scheme that realization:
A kind of computer readable storage medium, is stored thereon with computer program, and the computer program is held by processor
A kind of intelligent method for classifying based on thermal infrared imager as described in one of the object of the invention is realized when row.
Compared with prior art, the beneficial effects of the present invention are:
A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium of the invention, it is infrared by obtaining
Original uncorrected data, processing obtains the wide pixel collection of corresponding dynamic object object and the wide pixel of corresponding static object object respectively
Set handles to obtain two polarization datas of object by mapping and second level, compares dynamic instrumentation template library, static spy respectively
Targets of type can be judged by surveying template library, realize the intelligent recognition and classification of object.Simultaneously not by light luminance, shade
The influence of equal complex environments meets the object detection classification of varying environment.
Detailed description of the invention
Fig. 1 is to invent a kind of intelligent method for classifying flow chart based on thermal infrared imager；
Fig. 2 is to invent a kind of intelligent classification system connection figure based on thermal infrared imager；
Fig. 3 is Fig. 2 system execution flow chart；
Fig. 4 is rectangle data block schematic diagram.
Specific embodiment
In the following, being described further in conjunction with attached drawing and specific embodiment to the present invention, it should be noted that not
Under the premise of conflicting, new implementation can be formed between various embodiments described below or between each technical characteristic in any combination
Example.
Embodiment one:
Embodiment one discloses a kind of intelligent method for classifying based on thermal infrared imager, as shown in Figure 1, including following step
It is rapid:
Bis- polarization process step of S1 obtains infrared original uncorrected data, counts each object contour pixel point set, pass through
Mapping and two polarization operations obtain two polarization datas of corresponding dynamic object object and two polarization datas of corresponding static object object；
S2 template library compares step, by two polarization datas of corresponding dynamic object object, compares dynamic instrumentation template library, judgement
Targets of type；By two polarization datas of corresponding static object object, static instrumentation template library is compared, judges targets of type.
The above method, intelligent classification system such as Fig. 2 institute are realized using an intelligent classification system based on thermal infrared imager
Show comprising thermal infrared imager and pc client or cell phone client for receiving result of the investigations, wherein thermal infrared imager is again
Including infrared detector and infrared FPGA module.Data acquisition and data point of the thermal infrared imager as target acquisition categorizing system
The means of analysis constantly obtain the infrared original uncorrected data in thermal infrared imager field range, and pass through its internal intelligent mesh
Mark classification method is quickly handled and is analyzed to data, outlines the object detected in real time on infrared video stream picture
Position, and the information such as the type of the target detected, quantity are sent to pc client and cell phone client, after client receives
The result of detection and statistic of classification is shown on interface.It should be noted that the intelligence based on thermal infrared imager of the present embodiment
Classification method generally only needs thermal infrared imager and client that the intelligent detecting and classification method of target can be realized, preferably just
Case while using thermal infrared imager and client can also means by high-definition camera as supplementary observation, by cloud
Platform facilitates the operation and use of user.Thermal infrared imager is in use, by thermal infrared imager against needing to carry out Intelligent target spy
It surveys and the scene location of classification, opening intelligent detecting switch, thermal infrared imager starts that the scene in visual field is detected and divided
Analysis.The infrared original uncorrected data in field range is continuously obtained by infrared detector.Then the FPGA of infrared equipment
The systematic Intelligent target detection of inside modules and statistic of classification algoritic module, analyze the data got accordingly,
Finally analyze goal-selling object in the scanning scene type (include but are not limited to people, electric vehicle, car, bus and
Truck etc.), the coordinate position of the central pixel point of quantity and each target.Thermal infrared imager passes through the target species analyzed
Class and center point coordinate position come out various target real-time marks using different labels on infrared video stream.Finally
It can receive the detection data of equipment by pc client and cell phone client, and shown on the interface of client.
User can pass through the video flowing real time inspection of thermal infrared imager to the target and its distribution in infrared visual field in client simultaneously
Situation.
It is described below with reference to intelligent method for classifying process based on thermal infrared imager of the Fig. 3 to the present embodiment:
For thermal infrared imager normally after upper electricity operation, adjusting its angle makes it against the scene location detected.It beats
Client is opened by network connection thermal infrared imager, and obtains infrared video stream, clicks the button for starting intelligent detecting, it is infrared i.e.
It can start intelligent detecting and discriminance analysis that target is carried out to the scene location.The FPGA module of infrared inside begins through infrared
Detector constantly obtains infrared original uncorrected data, and substitutes into its internal Intelligent target probe algorithm and analyzed.
The Intelligent target detection sorting algorithm of system includes dynamic instrumentation sorting algorithm and two kinds of static instrumentation sorting algorithm
It constitutes, dynamic investigates algorithm for detecting the object moved under the scene, and static instrumentation algorithm is quiet under the scene for investigating
Only motionless object, type, quantity and the target detected of final output object are on thermal infrared imager picture
Coordinate position.The dynamic instrumentation sorting algorithm and static instrumentation sorting algorithm of this system are real in the method for machine learning
Existing, this system has trained the target template library of dynamic instrumentation and two kinds of the target template library of static instrumentation, and system is constantly adopted
Collect the target data under all kinds of monitoring scenes, classify to collected all kinds of target datas, respectively by manually and automatically
Mode carry out the model training of different type target, ultimately form the template characteristic library of all kinds of detection targets.Because of template library
Quantity it is more, the object accuracy rate finally analyzed is higher, but the quantity of template library is more, comparative analysis get up expend money
Source and time are bigger, but because thermal infrared imager data advantage, this system can under conditions of guaranteeing accuracy rate,
The data of template library carry out mapping and two polarization process, the actual size of template library are greatly reduced, to greatly improve template
The comparative analysis efficiency in library.Dynamic instrumentation sorting algorithm and static instrumentation sorting algorithm include the training of target feature library model
Comparison recognition methods two parts of the feature database of method and object, wherein the training method of target feature library model includes system again
Count two polarization operations of each object pixel collection and mapping.
Wherein the training method of the feature database model of dynamic instrumentation sorting algorithm is as follows:
When thermal infrared imager face a scene when, each pixel of infrared image have a brightness value (or
For gray value), which is original uncorrected data, and most black pixel uncorrected data value is 0, most white pixel uncorrected data value
It is 55296.The value of the uncorrected data of each pixel of whole image is all distributed in 0 to 55296 section, more black pixel,
The value of uncorrected data is closer to 0, and whiter (both brighter) pixel, the value of uncorrected data is closer to 55296.Respectively with infrared
The top edge of image and left edge are X-axis and Y-axis, establish rectangular coordinate system, then each infrared pixel is in two
It ties up in coordinate system, each pixel corresponds to a two-dimensional coordinate.It is infrared when the every scanning of equipment is to a detecting scene
FPGA module can continuously obtain the infrared original uncorrected data of the infrared space of a whole page all pixels point under the scene, respectively preceding
The infrared data of two continuous frames is stored in two-dimensional array afterwards.If the resolution ratio of thermal infrared imager is 640*480, illustrate infrared heat
As instrument horizontal pixel point quantity be 640, longitudinal pixel quantity be 480, then create two two-dimensional arrays
PrePiexValue [480] [640] and curPiexValue [480] [640] stores the pixel of previous frame and this frame respectively
Data.When there is target to move in detection scene, the data value of the pixel of corresponding position in thermal infrared imager image layout
It can vary widely therewith, likewise, certain variation has occurred in the pixel data value of some position if finding, illustrate this
There are the objects of movement for the corresponding position of pixel.
Two frame data of front and back are compared according to the subscript of the two-dimensional array of infrared uncorrected data, i.e., compare two two-dimemsional numbers respectively
Group, respectively to the pixel number of two two-dimensional array prePiexValue [m] [n] and curPiexValue [m] [n] same index
It is made the difference one by one according to value, if being shown to find according to the lot of experimental data of this system, the data value of the pixel of some position occurs
The absolute value of the difference of variation more than 50, i.e. prePiexValue [m] [n] and curPiexValue [m] [n] is more than or equal to
50, just illustrate that the position has target to move, which is added to the pixel Array for structural body of variation
In VarPiexl [], pixel dot-patterned structure is defined as follows shown:
typedef Pixel
{
int pixel_x；The X-coordinate of // pixel
int pixel_y；The Y-coordinate of // pixel
int value；The uncorrected data value of // pixel
}
Pixel VarPiexl[]；
The pixel saved in Array for structural body VarPiexl [] is each moving target object institute in the infrared space of a whole page substantially
Partial pixel point in the pixel block accounted for set (because when object moves in infrared picture, object
The data variation of edge pixel point can be more obvious, so centainly including each moving target object profile in the pixel collection
Pixel).The two-dimensional coordinate system created according to front, if certain two pixels position on infrared picture is adjacent, two pictures
The coordinate (X1, Y1) of vegetarian refreshments and (X2, Y2) can meet (- 2 < X1-X2 < 2) and (- 2 < Y1-Y2 < 2) two conditions simultaneously, creation
Pixel in VarPiexl [] array is carried out piecemeal processing (i.e. number by the Array for structural body OutLine [] of Pixel type
Pixel adjacent to each other is divided into one piece in group), each point of good pixel block is each target of infrared version in-plane moving
The profile information of each object, is stored in Array for structural body OutLine [] respectively by the pixel collection information of object profile
In.Because each object is to be made of in the infrared space of a whole page the set of several pixels, for each target
Block traverses the pixel array set of its profile information, find out the object block it is most upper, most under, most left and most right pixel
Point position, that is, find out maximum x coordinate MaxX, the maximum y-coordinate in the object contour pixel point array all pixels point
MaxY, the smallest x coordinate MinX and the smallest y-coordinate MinY.According to top left co-ordinate (MinX, MinY) and bottom right angular coordinate
Two coordinates of (MaxX, MaxY) intercept out the complete rectangle data block of the object, at this with the two points for diagonal line
In rectangular extent, the data in object block profile are still filled with the infrared data of the place space of a whole page of object block, outside number
It is filled according to 0, for rectangle data block schematic diagram as shown in figure 4, black is the pixel number evidence of object, white is according to right
The pixel that angular coordinate is expanded, value are filled with 0, and the data of the rectangular block are stored in ObjValue [] array.
After intercepting out object tile data, need to carry out data mapping and two polarization process.Mapping and two polarization
Algorithm, the data block (ObjValue [] array) of object can be mapped in the two-dimensional matrix of 16*16, and data
Carry out two polarization process.Mapping and two polarized algorithms are accomplished by
The process of mapping, be when pixel quantity of the object shared by x-axis direction be greater than/less than 16 or object is in y
Pixel quantity shared by axis direction be greater than/less than 16 when, need the two-dimensional matrix pressure the pixel number evidence where object
Contract/be extended to 16*16 pixel number according to the matrix constituted.In mapping process, algorithm first finds out naked number in object matrix
According to the value of the maximum pixel of value and the smallest pixel of uncorrected data value, maximum value MaxValue and minimum value are then calculated
The difference DiffValue of MinValue, while acquiring the average value AveValue of all pixels point of the object matrix.
If the value of some pixel mappedValue [i] [j] in the mapping matrix of 16*16 is respectively as follows:
For (int j=0；j<16；j++)
{
For (int i=0；i<16；i++)
{
MappedValue [i] [j]=GetAve (MinX+j* ((MaxX-MinX)/16), MinY+j* ((MaxY-Min
Y)/16),MinX+(j+1)*((MaxX-MinX)/16),MinY+(j+1)*((MaxY-MinY)/16))；
}
}
Wherein GetAve () function is the average value for obtaining several pixel gray values for mapping.It is defined as follows institute
Show:
By above-mentioned algorithm, the pixel rectangular block of object can be mapped in the data matrix of 16*16.Then right
Pixel number in the matrix of 16*16 is according to carrying out two polarization process, and when two polarization process is divided into two kinds, and one is be based on object
Two polarization of shape, another kind are two polarization based on object high/low temperature.
Wherein, based on two polarization of object shape: through the pixel collection of above-mentioned object profile, within profile
The value of the corresponding mapping matrix of pixel be all set to 1, the value the corresponding mapping matrix of pixel other than profile is whole
It is set to 0.At this time the value of pixel shared by object is all 1 in the matrix of 16*16, and around object other than pixel
Value is all 0, as the polarization of object shape two matrix.
Two polarization based on object high/low temperature: first acquiring the average value of data in matrix, then that data in matrix are big
It is denoted as 1 in the data of average value, the data that average value is less than in matrix are denoted as 0.Implementation method is as follows:
Training dynamic instrumentation template library when, to moving target object different types of in the various scenes of infrared acquisition (including
But it is not limited only to people, electric vehicle, car, bus and truck etc.) data handle according to above-mentioned steps, first pass through template library
Comparison recognition methods carry out automatic comparison identification.When carrying out system automatic identification using the comparison recognition methods of template library: being
System passes through two polarization datas for constantly counting each object block respectively according to above method to infrared original uncorrected data, with
Two polarization characteristic data of all kinds of targets in the dynamic instrumentation template library that front is added compare, and pass through shape feature library and height
Low temperature characteristics library compares respectively, the data after comparison meet shape feature library percent 90 if determine the object category
In such target, or meet 60 the percent of shape feature library, while meeting 75 the percent of high/low temperature feature database,
It can determine that the object belongs to the category.
There are also the functions that storage storage manually is carried out to incorrect matched object simultaneously for this system, to various objects
Categorical data stored.If after the comparison of dynamic instrumentation template library, some object detected mismatches dynamic template
Any targets of type in library then triggers abnormal object alarm, the data the abnormal object and the position on infrared picture
It sets and preserves, user can be by actually looking at the type of target, which kind of type dynamic analog which is added to by selection
Version library, if the target type is not belonging to existing object library type, user can create dynamic object template types library, then
Two polarization data of rectangle of the target is saved into newly-built library, it, can be certainly if identical targets of type occurs again in next time
The dynamic data with the object library compare.During the comparison process, position, the type for recording each object, count simultaneously
The quantity of various types object out, while being carried out according to the type of each target using different labels on infrared video stream
Real-time calibration.After the completion of all objects detected are all analyzed, continue to obtain next frame infrared data, the frame with front
Infrared data continues analysis processing according to above step.This method is special with the template of all kinds of targets in dynamic template library
Sign the included data volume in library is continuously increased, and system also can be higher and higher to the discrimination of the target of various species.
The training method of the feature database model of static instrumentation sorting algorithm is as follows:
Similar with dynamic instrumentation sorting algorithm, static instrumentation sorting algorithm also needs first to find out shared by object first
Pixel data block, this system screen object block when, first pass through image processing algorithm and extract each object block profile
Pixel, then the data of each object block are extracted, finally differentiate whether object block belongs to detected target type again.
Firstly, carrying out edge extracting to the image of infrared data: at present for the algorithm of Edge extraction very at
It is ripe, the operator of common Edge extraction have Sobel operator, Isotropic Sobel operator, Roberts operator,
Prewitt operator, Laplacian operator and Canny operator pass through the comparison of the resultant effect to above-mentioned several operators, Canny
The effect of operator is more several than front will be good, and this system is realized when carrying out object block edge extracting using Canny operator.
Canny operator, which is one, has filtering, enhances, the multistage Optimizing operator of detection, before being handled,
Canny operator carrys out smoothed image first with Gaussian filter with except denoising, Canny partitioning algorithm is using single order local derviation
Finite difference calculates gradient magnitude and direction, and during processing, Canny operator will also be by non-maxima suppression
Process, last Canny operator also use two threshold values to connect edge.
Canny edge detection algorithm is divided into 4 steps: step1: using Gaussian filter smoothing image；Step2: with one
The finite difference of rank local derviation calculates amplitude and the direction of gradient；Step3: non-maxima suppression is carried out to gradient magnitude；
Step4: edge is detected and connected with dual threashold value-based algorithm.Canny edge detection algorithm is more conventional, following is a brief introduction of lower and is
Some details that system uses.
Step1: it is filtered with Gaussian filter and realizes data smoothing processing
Realize smoothed image usually using two-dimensional Gaussian function using Gaussian filter:
Wherein σ is the standard deviation of Gaussian function, and x and y are the coordinate of Gaussian template, x2+y2It can be appreciated that the point of filtering
Square with a distance from template center's point.
According to the principle, it can realize that the generating function method of Gaussian template matrix is as follows with code:
Wherein first parameter gaus is directed to the pointer of the Gaussian template generated, and second parameter size is Gaussian convolution
The size of core；Third parameter sigma is the standard deviation of convolution kernel.By the comparison of actual treatment, matrix size is used
For 5*5, the Canny operator matrix that convolution kernel standard deviation sigma is 1, i.e., in above-mentioned function, parameter is that size is 5, sigma 1
It substitutes into and calculates, get Gaussian template matrix.The Gaussian template matrix and infrared data of above-mentioned acquisition are subjected to convolution, pressed down
The infrared data after gaussian filtering after noise processed.
Step2: amplitude and the direction of gradient are calculated with the finite difference of single order local derviation
The gradient of gray value of image generally carries out approximation using first difference point, can thus obtain image in x and y
Two matrixes of partial derivative on direction, this system calculate gradient magnitude and direction using common Sobel operator, and operator uses warp
The Sobel convolution operator of allusion quotation: the gradient operator on the direction x and the direction y is respectively
X-direction operator is as follows:
| -1 | 0 | +1 |
| -2 | 0 | +2 |
| -1 | 0 | +1 |
Y-direction operator is as follows:
| +1 | +2 | +1 |
| 0 | 0 | 0 |
| -1 | -2 | -1 |
Convolution Formula in X-direction and Y-direction are as follows:
Wherein I is to acquire horizontal direction respectively (i.e. using above-mentioned Convolution Formula by previous step treated infrared data
The direction x) and vertical direction (i.e. the direction y) on convolved data, then again to the gradient data both horizontally and vertically acquired
Each point sum, i.e.,
G is to acquire gradient data, while calculating the value of the gradient angle of each point, i.e.,
θ=a tan (Gy/Gx)*57.3+90
Step3: non-maxima suppression is carried out to gradient magnitude, further eliminates non-edge noise, while narrowing edge picture
The width of vegetarian refreshments.
System respectively with the value of θ 0 to 45 degree, 45 to 90 degree, 90 to 135 degree and 135 to four angular ranges of 180 degree into
The non-maximum processing of row, the amplitude formula of each point are as follows:
P(X0, Y0)+(P(X0- 1, Y0+1)-P(X0, Y0))*tan(θ)
The gradient data and gradient angle for each point that third step above is acquired are in four angular ranges respectively using upper
It states formula to be handled, obtains the infrared data of non-maxima suppression.
Step4: being detected with dual threashold value-based algorithm and connection edge, dual threshold one Low threshold A and a high threshold B, generally
Taking B is the 70% of image overall data Distribution value, and B is the A of 1.5 to 2 times of sizes；And data value is set to 255, ash greater than B
Angle value is set to 0 less than A, and whether data value has 255 value between A and B in checking surrounding 8 points, if so,
Then the point is also set to 255, if not having, the value of the point is set to 0.By this system referring to the real data situation of thermal infrared imager
Comparison, finally taking the value of B is 121, and taking the value of A is 72.
The above-mentioned realization principle for Canny operator, the library opencv is classical image processing algorithm, wherein including
The realization of Canny operator, this system call directly cvCanny function therein to realize the extraction to objective contour:
void cvCanny(const CvArr*image,CvArr*edges,double threshold1,double
Threshold2, int aperture_size=3)；
Parameter declaration: image is the image data of input；Edges is the image border data of output；Threshold1
One threshold value；Second threshold value of threshold2；Aperture_size Sobel operator kernel size.
By above-mentioned processing, infrared picture data after can be processed recalls cvFindContours function
To extract the pixel coordinate set in each objective contour:
int cvFindContours(CvArr*image,CvMemStorage*storage,CvSeq**first_
contour,
Int header_size=sizeof (CvContour), int mode=CV_RETR_LIST,
Int method=CV_CHAIN_APPROX_SIMPLE, CvPoint offset=cvPoint (0,0))；
Parameter declaration: image is the image data after binaryzation；Storage is the container of the profile extracted；first_
Contour is the pointer for being directed toward first external profile；Header_size is the size of profile sequence；Mode is profile retrieval
Hierarchical schema；Method refers to the approximate method in edge；Offset refers to offset；
Parameter mode selects CV_RETR_LIST when this system is called, and retrieves all profiles, and put it into list；
Method selects CV_CHAIN_APPROX_SIMPLE；Offset selects default value, no offset.It can be obtained using above-mentioned function
The wire-frame image vegetarian refreshments coordinate set for getting each object block in image is stored in the container of storage pointer direction.
After getting the pixel coordinate set of each profile, system be directed to respectively each object block contour area into
Row analysis.If the untreated infrared original uncorrected data collection of the frame is combined into M (x, y), after treatment, wherein some object block profile
Pixel data acquisition system be P (x, y), the pixel in the profile set is ranked up from small to large according to Y-coordinate, then looks for
The smallest X-coordinate Xmin and maximum X-coordinate Xmax in every a line (i.e. identical Y value) out is then indulged in the frame data set and is sat
It is designated as Y and the data of pixel of the abscissa between Xmin and Xmax is added in the container of the object block data, in this way
Just the data of each object block in the infrared original uncorrected data of the frame are all extracted respectively.Each object block is counted simultaneously
Pixel number, object block pixel minimum x coordinate minX, maximum x coordinate maxX, minimum y-coordinate minY and maximum y-coordinate
MaxY, and its shared pixel number on the direction x and the direction y.Then prescreening, prescreening are carried out to the data of object block
When, system gets the general distance parameter of equipment distance monitoring position under the presetting bit of user setting, according to this system
The ratio formula of distance and pixel calculates the usable range of object block pixel.Wherein the fixed pixel point ratio of system with
Distance statistics curve equation:
F (x)=8.61 × 10-14x5-2.58×10-10x4+2.86×10-7x3+0.02x-0.11
Wherein F (x) is the length (unit: rice) of the corresponding actual object of 10 pixel length, and x is object and front end
The distance between equipment (unit: rice),
Such as calculating the corresponding realistic objective object length of 10 infrared image vegetarian refreshments in the presetting bit scene is 2 meters, if this
It is about 6 meters that scape, which needs the maximum physical length of object detected, and width is about 2 meters, then the object is in infrared upper length
Pixel number shared by degree reality should be within 30, and pixel number shared by width reality should be within 10, then exists
It is got rid of when prescreening, the length and width of object block and is gone in the same way considerably beyond the object block data of the range
Remove the object block data that object block pixel number is far smaller than the range.If equipment distance under the not set presetting bit of user
The distance parameter of position is monitored, system can detect the target vehicle institute under the presetting bit by the dynamic instrumentation algorithm of front
The quantity of the pixel accounted for, such as detecting car pixel shared in the presetting bit is 12*6, and the actual length of car
It is about 4 meters * 2 meters with width, then calculates under the presetting bit, pixel number ratio shared by one meter is 3, then with the ratio
Multiplied by the physical length for the maximum object block for needing to detect, the model of the maximum object block under the presetting bit can be calculated
It encloses.Similarly, by prescreening, those shared pixel quantities are got rid of not between the minimum zone and maximum magnitude
Object block data.
Object block pixel number is filtered out after, as dynamic instrumentation algorithm, respectively for the target each detected
The data area of block is extended for a rectangular extent, and the diagonal line coordinates of rectangular extent is (minX, minY) and (maxX, maxY),
In the rectangular extent, the data in object block profile are still filled with the data of object block, outside data with 0 filling.
When the data of each object block are after being expanded into rectangular extent, again according to reflecting in above-mentioned dynamic instrumentation algorithm
It penetrates and is handled with two polarized methods, but only carry out two polarization process of shape herein.
When training static instrumentation template library, to different types of object data in the various scenes of infrared acquisition according to upper
The step of face, is handled, and the comparison recognition methods for first passing through template library carries out automatic comparison identification.
When carrying out system automatic identification using the comparison recognition methods of template library: system obtains two polarization of each object block
After data, the two polarization characteristic data of all kinds of targets in static instrumentation template library added with front are compared, if when comparison
The data of the target are identical as percent 90 of some two polarization matrix data in feature database, it is determined that the object belongs to such
Target.There are also the functions that storage storage manually is carried out to incorrect matched object simultaneously for this system, to various objects
Categorical data is stored.If after the comparison of static instrumentation template library, some object detected mismatches static template library
In any targets of type, then the alarm of Exceptional static target is triggered, the data of the abnormal object and on infrared picture
Position preserves, and user can be by actually looking at the type of target, which kind of type static which is added to by selection
Template library, if the target type is not belonging to existing target type in object library, user can create static object class pattern
Then plate library is saved into two polarization data of the rectangle of the target in newly-built library, if identical object occurs again in next time
Type can be compared with the data of the object library automatically.During the comparison process, position, the type of each object are recorded,
The quantity of various types object is counted simultaneously, while being used according to the type of each target on infrared video stream different
Label carries out real-time calibration.
After the completion of all objects detected are all analyzed, continue to obtain next frame infrared data according to above step
Continue analysis processing.This method in dynamic template library the included data volume in template characteristic library of all kinds of targets it is continuous
Increase, system also can be higher and higher to the discrimination of the target of various species.Because two polarization after feature database data with two into
The form of file processed saves as feature library file, substantially reduces the size of object characteristic data file, to greatly reduce
The density of feature database valid data, not only increases the reading of feature database data and to specific efficiency, ensure that real-time analysis
Under conditions of, while also considerably increasing the discrimination of object.It is visited simultaneously when each frame infrared data carries out static and dynamic
Then the type for analyzing each object respectively after survey and the coordinate position on the infrared space of a whole page are superimposed by OSD,
Real-time mark is carried out to target on infrared video stream, while counting the quantity of various types object, then statistical information
It is sent to client, client shows the statistical information of targets of type and quantity on interface.The statistics of object
Information can be stored in MySQL database according to the time, and user can also inquire the object letter counted in each period
The report data of breath.
Embodiment two:
Embodiment two discloses a kind of readable computer storage medium, which is somebody's turn to do for storing program
When program is executed by processor, a kind of intelligent method for classifying based on thermal infrared imager of embodiment one is realized.
The above embodiment is only the preferred embodiment of the present invention, and the scope of protection of the present invention is not limited thereto,
The variation and replacement for any unsubstantiality that those skilled in the art is done on the basis of the present invention belong to institute of the present invention
Claimed range.
Claims (10)
1. a kind of intelligent method for classifying based on thermal infrared imager characterized by comprising
Two polarization process steps obtain infrared original uncorrected data, count each object contour pixel point set, by mapping and
Two polarization operations obtain two polarization datas of corresponding dynamic object object and two polarization datas of corresponding static object object；
Template library compares step, by two polarization datas of corresponding dynamic object object, compares dynamic instrumentation template library, judges object
Type；By two polarization datas of corresponding static object object, static instrumentation template library is compared, judges targets of type.
2. as described in claim 1 based on the intelligent method for classifying of thermal infrared imager, it is characterised in that: at two polarization
It manages in step, to the pixel collection of dynamic object object, marks off dynamic object object contour pixel point set, root after piecemeal processing
Object tile data is intercepted out according to rectangular edges pixel position, carries out mapping and two polarization operations.
3. as claimed in claim 2 based on the intelligent method for classifying of thermal infrared imager, it is characterised in that: at two polarization
It manages in step, object tile data is mapped in pixel matrix and carries out two polarization operations.
4. as claimed in claim 3 based on the intelligent method for classifying of thermal infrared imager, it is characterised in that: at two polarization
It manages in step, two polarization operations of corresponding dynamic object object include two polarization and two poles of object high/low temperature of object shape
Change.
5. as described in claim 1 based on the intelligent method for classifying of thermal infrared imager, it is characterised in that: at two polarization
It manages in step, for static object object, obtains corresponding static object by carrying out edge extracting to infrared original uncorrected data image
The contour pixel point set of object, by carrying out two polarization operations after prescreening.
6. as claimed in claim 5 based on the intelligent method for classifying of thermal infrared imager, it is characterised in that: at two polarization
It manages in step, two polarization operations of static object object are the polarization of shape two.
7. the intelligent method for classifying based on thermal infrared imager as claimed in any one of claims 1 to 6, it is characterised in that: in institute
It states template library to compare in step, if two polarization datas of dynamic object object mismatch object any in dynamic instrumentation template library
Type then triggers abnormal object alarm.
8. the intelligent method for classifying based on thermal infrared imager as claimed in any one of claims 1 to 6, it is characterised in that: in institute
It states template library to compare in step, when judging targets of type, while recording the position of object, be superimposed by OSD, infrared
Real-time mark is carried out to target on video flowing, and is uploaded to client after counting the quantity of corresponding object.
9. a kind of intelligent classification system based on thermal infrared imager, it is characterised in that: including infrared detector and FPGA module, institute
Infrared detector is stated for obtaining the infrared original uncorrected data in investigation field angle, the FPGA module and the infrared detector
A kind of intelligent method for classifying based on thermal infrared imager as described in claim 1-8 any one is realized in communication connection.
10. a kind of computer readable storage medium, is stored thereon with computer program, it is characterised in that: the computer program
A kind of intelligent classification side based on thermal infrared imager as described in claim 1-8 any one is realized when being executed by processor
Method.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811444291.8A CN109815784A (en) | 2018-11-29 | 2018-11-29 | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201811444291.8A CN109815784A (en) | 2018-11-29 | 2018-11-29 | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109815784A true CN109815784A (en) | 2019-05-28 |
Family
ID=66601589
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201811444291.8A Pending CN109815784A (en) | 2018-11-29 | 2018-11-29 | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109815784A (en) |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110189499A (en) * | 2019-06-20 | 2019-08-30 | 深圳市奥闻科技有限公司 | A kind of highly sensitive adaptive interior living body biological monitoring intelligence system |
| CN110610593A (en) * | 2019-10-16 | 2019-12-24 | 徐州筑之邦工程机械有限公司 | Intelligent safety early warning system for mining truck |
| CN111540166A (en) * | 2020-05-09 | 2020-08-14 | 重庆工程学院 | Unmanned aerial vehicle night search system and method based on deep learning |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101882217A (en) * | 2010-02-26 | 2010-11-10 | 杭州海康威视软件有限公司 | Target classification method of video image and device |
| CN101898567A (en) * | 2010-04-07 | 2010-12-01 | 西南交通大学 | Railway foreign body limit-intruding monitoring system based on intelligent video |
| CN202956192U (en) * | 2012-03-16 | 2013-05-29 | 东莞华仪仪表科技有限公司 | Device and system of intelligent thermal infrared imager |
| CN104299354A (en) * | 2014-10-28 | 2015-01-21 | 王翠萍 | Infrared thermal imaging hot spot detection alarm device and method |
| CN105404857A (en) * | 2015-11-04 | 2016-03-16 | 北京联合大学 | Infrared-based night intelligent vehicle front pedestrian detection method |
| EP3035239A1 (en) * | 2014-12-02 | 2016-06-22 | Xerox Corporation | Adapted vocabularies for matching image signatures with fisher vectors |
| CN106131499A (en) * | 2016-07-28 | 2016-11-16 | 广州紫川电子科技有限公司 | Same monitoring position based on thermal infrared imager multizone monitoring method, Apparatus and system |
| CN106600628A (en) * | 2016-12-13 | 2017-04-26 | 广州紫川电子科技有限公司 | Target object identification method and device based on infrared thermal imaging system |
| CN107295230A (en) * | 2017-07-14 | 2017-10-24 | 广州紫川电子科技有限公司 | A kind of miniature object movement detection device and method based on thermal infrared imager |
| CN107707810A (en) * | 2017-08-21 | 2018-02-16 | 广州紫川电子科技有限公司 | Thermal source method for tracing, apparatus and system based on thermal infrared imager |
| CN107728160A (en) * | 2017-08-21 | 2018-02-23 | 广州紫川电子科技有限公司 | A kind of vision radar detection method and system based on thermal infrared imager |
| CN108414091A (en) * | 2018-02-09 | 2018-08-17 | 广州紫川电子科技有限公司 | A kind of contactless temperature-measuring warning device for petrochemical equipment |
-
2018
- 2018-11-29 CN CN201811444291.8A patent/CN109815784A/en active Pending
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101882217A (en) * | 2010-02-26 | 2010-11-10 | 杭州海康威视软件有限公司 | Target classification method of video image and device |
| CN101898567A (en) * | 2010-04-07 | 2010-12-01 | 西南交通大学 | Railway foreign body limit-intruding monitoring system based on intelligent video |
| CN202956192U (en) * | 2012-03-16 | 2013-05-29 | 东莞华仪仪表科技有限公司 | Device and system of intelligent thermal infrared imager |
| CN104299354A (en) * | 2014-10-28 | 2015-01-21 | 王翠萍 | Infrared thermal imaging hot spot detection alarm device and method |
| EP3035239A1 (en) * | 2014-12-02 | 2016-06-22 | Xerox Corporation | Adapted vocabularies for matching image signatures with fisher vectors |
| CN105404857A (en) * | 2015-11-04 | 2016-03-16 | 北京联合大学 | Infrared-based night intelligent vehicle front pedestrian detection method |
| CN106131499A (en) * | 2016-07-28 | 2016-11-16 | 广州紫川电子科技有限公司 | Same monitoring position based on thermal infrared imager multizone monitoring method, Apparatus and system |
| CN106600628A (en) * | 2016-12-13 | 2017-04-26 | 广州紫川电子科技有限公司 | Target object identification method and device based on infrared thermal imaging system |
| CN107295230A (en) * | 2017-07-14 | 2017-10-24 | 广州紫川电子科技有限公司 | A kind of miniature object movement detection device and method based on thermal infrared imager |
| CN107707810A (en) * | 2017-08-21 | 2018-02-16 | 广州紫川电子科技有限公司 | Thermal source method for tracing, apparatus and system based on thermal infrared imager |
| CN107728160A (en) * | 2017-08-21 | 2018-02-23 | 广州紫川电子科技有限公司 | A kind of vision radar detection method and system based on thermal infrared imager |
| CN108414091A (en) * | 2018-02-09 | 2018-08-17 | 广州紫川电子科技有限公司 | A kind of contactless temperature-measuring warning device for petrochemical equipment |
Non-Patent Citations (2)
| Title |
|---|
| 胡明春等著: "《雷达目标识别原理与试验技术》", 31 December 2017, 国防工业出版社 * |
| 黄萍等: "基于MLX90621红外传感器的室内人数检测", 《电子测量技术》 * |
Cited By (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110189499A (en) * | 2019-06-20 | 2019-08-30 | 深圳市奥闻科技有限公司 | A kind of highly sensitive adaptive interior living body biological monitoring intelligence system |
| CN110610593A (en) * | 2019-10-16 | 2019-12-24 | 徐州筑之邦工程机械有限公司 | Intelligent safety early warning system for mining truck |
| CN111540166A (en) * | 2020-05-09 | 2020-08-14 | 重庆工程学院 | Unmanned aerial vehicle night search system and method based on deep learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN108038476B (en) | A kind of facial expression recognition feature extracting method based on edge detection and SIFT | |
| CN106856002B (en) | Unmanned aerial vehicle shooting image quality evaluation method | |
| Huang et al. | A new building extraction postprocessing framework for high-spatial-resolution remote-sensing imagery | |
| CN103984961B (en) | A kind of image detecting method for being used to detect underbody foreign matter | |
| CN107909081B (en) | Method for quickly acquiring and quickly calibrating image data set in deep learning | |
| CN108985169B (en) | Shop cross-door operation detection method based on deep learning target detection and dynamic background modeling | |
| CN108573276A (en) | A kind of change detecting method based on high-resolution remote sensing image | |
| CN111079556A (en) | Multi-temporal unmanned aerial vehicle video image change area detection and classification method | |
| CN106530281A (en) | Edge feature-based unmanned aerial vehicle image blur judgment method and system | |
| CN105069818A (en) | Image-analysis-based skin pore identification method | |
| CN106067026A (en) | A kind of Feature extraction and recognition search method of microimage of Chinese medical herb | |
| CN110189375A (en) | A kind of images steganalysis method based on monocular vision measurement | |
| CN105893946A (en) | Front face image detection method | |
| CN110335233B (en) | Highway guardrail plate defect detection system and method based on image processing technology | |
| CN109815784A (en) | A kind of intelligent method for classifying based on thermal infrared imager, system and storage medium | |
| Ji et al. | In-field automatic detection of maize tassels using computer vision | |
| CN108765406A (en) | A kind of snow mountain detection method based on infrared remote sensing image | |
| CN110111283A (en) | The reminding method and system of infrared suspected target under a kind of complex background | |
| CN111665199A (en) | Wire and cable color detection and identification method based on machine vision | |
| WO2023231262A1 (en) | Hoisting wire rope tension measurement method based on visual vibration frequency identification | |
| Zhao et al. | An adaptation of CNN for small target detection in the infrared | |
| CN114863106B (en) | Centrifugal fertilizer spreading particle distribution detection method | |
| Ouma et al. | Urban features recognition and extraction from very-high resolution multi-spectral satellite imagery: a micro–macro texture determination and integration framework | |
| CN109886195A (en) | Skin identification method based on depth camera near-infrared single color gradation figure | |
| Jia et al. | Research on the decomposition and fusion method for the infrared and visible images based on the guided image filtering and Gaussian filter |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190528 |
|
| RJ01 | Rejection of invention patent application after publication |