CN111461011A - Weak and small target detection method based on probabilistic pipeline filtering - Google Patents
Weak and small target detection method based on probabilistic pipeline filtering Download PDFInfo
- Publication number
- CN111461011A CN111461011A CN202010248459.9A CN202010248459A CN111461011A CN 111461011 A CN111461011 A CN 111461011A CN 202010248459 A CN202010248459 A CN 202010248459A CN 111461011 A CN111461011 A CN 111461011A
- Authority
- CN
- China
- Prior art keywords
- target
- pipeline
- image
- frame
- candidate
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/26—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion
- G06V10/267—Segmentation of patterns in the image field; Cutting or merging of image elements to establish the pattern region, e.g. clustering-based techniques; Detection of occlusion by performing operations on regions, e.g. growing, shrinking or watersheds
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V2201/00—Indexing scheme relating to image or video recognition or understanding
- G06V2201/07—Target detection
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Artificial Intelligence (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Signal Processing (AREA)
- General Engineering & Computer Science (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a weak and small target detection method based on probabilistic pipeline filtering. The method mainly solves the problem that the existing small target detection technology is low in precision and robustness. The implementation scheme is as follows: firstly, preprocessing an input sequence image for background suppression, and obtaining a binary image by using a maximum inter-class variance method; then initializing pipeline parameters, and establishing a filtering pipeline by using the existing N frames of binary images; and then, detecting the preprocessed image by using probabilistic pipeline filtering, namely, eliminating easily distinguished noise points by space constraint, calculating the probability of the candidate target which belongs to a real target by using gray distribution, regional characteristics and the like for the candidate target which is difficult to distinguish, and outputting the candidate target with the maximum probability as the real target. Compared with the traditional pipeline filtering, the method effectively solves the problem that the class target noise interferes with a real target, improves the precision and robustness of target detection, reduces the omission factor and the false alarm rate, and can be used for detecting and tracking weak and small targets under a complex background.
Description
Technical Field
The invention belongs to the technical field of pattern recognition, and particularly relates to a target detection method which can be used for detecting and tracking weak and small targets under a complex background.
Background
The detection of weak and small targets is one of the hot problems in the field of photoelectric detection and tracking, the energy of the targets is far less than that of the background due to the imaging principle, the imaging distance and the like, and meanwhile, a large amount of noise interference is introduced due to the wide visual field. Due to the fact that the background is complex, the target is weak and is often submerged in noise, image contrast is low, the signal-to-noise ratio is high, and due to the fact that the occupation ratio of target pixels is small, the target is lack of pixel characteristics such as textures, and the characteristics which can be used for target detection are few.
At present, most of classical detection algorithms for weak and small targets are based on single-frame detection, but due to the fact that interference of a large number of similar targets exists in the background, satisfactory effects of single-frame detection are difficult to achieve, and if the detection accuracy is to be improved, correlation and continuity among multiple frames of targets need to be fully utilized. Due to the fact that noise is random, false targets can be removed well through multi-frame detection, and real target tracks are output. The pipeline filtering is a commonly used multi-frame detection algorithm, and liu man et al 2007 reported in the university of western's electronics science and technology "infrared small object detection based on mobile weighted pipeline filtering" in an algorithm that introduces weighted pipeline center coordinate displacement when modifying the pipeline center coordinate position. Xuzhiyong et al propose a pipe diameter adaptive pipeline filtering algorithm in a patent (patent number: CN106469313B) Weak target detection method of pipe diameter adaptive time-space domain filtering, which modifies the pipe diameter according to target scale change, but the algorithm cannot effectively overcome strong noise interference due to the fact that the consistency constraint of multi-frame target characteristics is not fully utilized, so that pipe diameter tracking errors occur, and further the robustness and precision of detection are influenced.
The methods only improve the method for updating the sizes of the center and the pipe diameter of the pipeline from the perspective of pipeline design, do not fully satisfy the correlation and continuity between multi-frame targets and characteristics thereof, cause low target detection precision and poor robustness, and cannot better adapt to the actual needs of the current engineering.
Disclosure of Invention
The invention aims to provide a weak and small target detection method based on probabilistic pipeline filtering aiming at the defects of the prior art, so that the accuracy and robustness of target detection are effectively improved by utilizing the correlation and continuity between multi-frame targets and characteristics thereof while the pipeline parameters are adaptively changed along with the target change, and the actual requirements of the current engineering are met.
In order to achieve the above purpose, the implementation steps of the invention comprise:
(1) the method comprises the steps of inputting an mth frame target image to be detected, wherein M is 1,2,., M, preprocessing the mth frame target image by using a RobinsonGuard spatial filter, and M is the total frame number of sequence images to be processed;
(2) performing target segmentation on the preprocessed input image by using a maximum inter-class variance method to obtain a binary image;
(3) building a pipeline: if the current input frame is a first frame, namely m is 1, defining the length of the pipeline as N and the pipe diameter as 2-3 times of the target size, copying N parts of binary images of the first frame to establish the pipeline, and jumping (6);
(4) and utilizing the established pipeline to perform target detection on the input image by using a probabilistic algorithm:
(4a) obtaining parameters of the size, the position and the characteristics of a corresponding target according to the existing N frames of images in the pipeline;
(4b) for the current input frame image, eliminating false targets which are not in the pipeline receptive field completely through space constraint;
(4c) using gray distribution constraint to further eliminate false target, i.e. calculating residual candidate of mth frameGray distribution feature vector of target: gamma rayt＝(maxt,mint,meant,vart,sumt) T1, 2, T, only remaining to satisfy dist (γ, γ)t)≤grayConditional candidate objects, where γ represents the gray-scale distribution feature vector of the real object, T represents the number of remaining candidate objects in the mth frame, maxt,mint,meant,vart,sumtRespectively represent the maximum, minimum, mean, variance and cumulative sum of the t-th candidate target gray-scale value,graya threshold value representing a gray distribution constraint;
(4d) calculating the credibility between the residual candidate targets in the mth frame and the real targets of the N frames of images in the pipeline: p is a radical oft＝[λ1dist(α,αt)+λ2dist(β,βt)]-1Obtaining the credibility of a plurality of candidate targets of the mth frame belonging to the real target, and carrying out data normalization processing on the credibility scores to obtain the probability q of the candidate targets belonging to the real targett＝softmax(pt) Wherein λ is1Representing the correlation coefficient, λ, between the true object and the candidate in terms of the shape characteristics of the object2Representing the correlation coefficient of the real target and the candidate target in terms of spatial features, dist (-) representing the distance between the two vectors, α and αtRepresenting the shape feature vector of the real object and the shape feature vector of the candidate object, β and β, respectivelytRespectively representing a real target space characteristic vector and a space characteristic vector of a candidate target;
(4f) output probability qtThe largest candidate target is used as the real target of the mth frame;
(5) updating the pipeline: updating the image to be processed in the pipeline by adopting a first-in first-out storage rule, namely moving out the image of the first frame in the pipeline, and moving the detection result of the mth frame into the pipeline finally;
(6) let m be m + 1: if M < M +1, returning to (1); otherwise, ending the target detection task and outputting M frames of target detection results.
Because the invention fully utilizes the known N frame target detection results to establish the filtering pipeline, compared with the prior method, the invention has the following advantages:
1) the invention updates the central coordinate of the pipeline in real time by using the position of the target, is beneficial to inhibiting the interference of edge noise on target detection and improves the signal-to-noise ratio in the reception field of the pipeline.
2) The pipeline reception field is automatically updated along with the change of the target size, namely the shape of the pipeline reception field is more flexible by using the target mask, and compared with the traditional pipeline filtering with fixed pipe diameter, the detection result is more accurate.
3) The method can fully utilize the continuity of the motion track of the real target and the correlation between the characteristics of the multi-frame target, improve the precision and the robustness of target detection and reduce the omission factor and the false alarm rate.
In the target detection image sequence, the target motion track has continuity, the correlation exists between the characteristics of continuous frame targets, and the noise has randomness, so that the correlation between a plurality of candidate targets and a real target in a current frame in a characteristic space can be calculated by constructing a method for embedding a target characteristic vector into a space, the probability that the candidate targets belong to the real target is obtained, and the candidate target with the maximum probability is output as the real target of the current frame.
Drawings
FIG. 1 is an overall flow chart of a probabilistic pipeline filtering algorithm of the present invention;
FIG. 2 is a 15 th frame detection process and result chart of the present invention for a small target sequence in a scene with multiple strong noise points;
FIG. 3 is a diagram showing the detection results of all frames of a small target sequence in a scene with multiple noise points according to the present invention;
FIG. 4 is a diagram of the detection process and results of the 15 th frame of the ground background small target sequence according to the present invention;
fig. 5 is a schematic diagram of the detection results of all frames of a ground background small target sequence according to the present invention.
Detailed Description
The invention establishes the filtering pipeline by fully utilizing the target information of the existing frame, not only can lead the pipeline receptive field to change in a self-adaptive way along with the change of the target size, but also can fully utilize the continuity of the real target motion track and the correlation among the characteristics of the multi-frame target, thereby effectively solving the interference of class target noise on the real target, improving the target detection precision and robustness and reducing the omission ratio and the false alarm ratio.
The following further describes embodiments of the present invention with reference to the drawings.
Referring to fig. 1, the implementation steps of this example are as follows:
step 1, inputting an mth frame target image to be detected, wherein M is 1, 2.
The Robinson Guard filtering is a nonlinear nonparametric filter, realizes background suppression by comparing the difference between the extreme values of a central pixel and the extreme values of adjacent pixels of the central pixel, and effectively retains complete information of weak and small targets by setting a Guard band based on template operation. The specific implementation of this step is as follows:
1.1) inputting a target image, and comparing the difference between the extreme values of a central pixel and adjacent pixels:
take the example of a filter template Z of size 7 × 7.
In the table, numbers a1 to a24 represent the outermost region of the filter template, i.e., the background region around the object.
For any pixel X on the input image, pass through with its surrounding pixels AiThe filtering of the pixel X is realized by maximum value comparison, and the specific formula is as follows:
whereinX pixels to be detected, AiI is 1,2, 24, which is the corresponding position pixel value of the input image;
1.2) setting the guard band to effectively reserve the weak target:
assuming that the target is an extremum signal of a local area, the extremum of the pixels outside the guard band is used to determine whether the central point X is the extremum signal of the local area, so as to distinguish the target pixel from noise, the target pixel is enhanced, and the weak and small target is maintained. That is, the Robinson guard filter compares the gray value X of the pixel to be processed with the gray extremum of the outermost region of the filtering template, and sets the window size of the filter suitable for the target size, that is, the width of the guard band, so that when the target size is equivalent to the size of the guard band, the pixels inside and at the edge of the target pixel are better maintained, and meanwhile, the noise smaller than the window size can be filtered, so as to obtain the pre-processed noise-reduced image, as shown in fig. 2(a) and fig. 4(a), where fig. 2(a) is a schematic diagram of the Robinson guard filtering result of the 15 th frame of the small target sequence in the multiple noisy-point scene in this example, and fig. 4(a) is a schematic diagram of the Robinson guard filtering result of the 15 th frame of the small target sequence in the ground background in this example.
And 2, performing target segmentation on the noise reduction image obtained by preprocessing to obtain a binary image.
The step is carried out by a maximum inter-class variance method, which is realized as follows:
2.1) calculating the ratio of the target pixel and the background pixel in the whole image:
assuming that the size of the grayscale picture I is W × H, the partition threshold of the foreground and the background is thr, setting the pixels with the grayscale values larger than the partition threshold as the foreground, and setting the pixels smaller than the partition threshold as the background, obtaining a binary image BW, wherein the value of the ith pixel in the binary image is represented as:
respectively calculating the proportion p of the foreground pixels in the whole image according to the parameters1And mean value mu of foreground pixels1：
Wherein ". x" denotes the multiplication of matrices by elements;
according to the proportion p of the foreground pixel to the whole image1And obtaining the proportion of the background pixel in the whole image as follows: p is a radical of0＝1-p1(ii) a Calculating the overall image mean value mu as:
2.2) calculating the between-class variance of the target and the background:
calculating the variance Vc between the foreground and the background according to the mean value and the proportion of the foreground pixel and the background pixel and the mean value of the whole image, wherein the formula is as follows:
Vc＝p0×(μ-μ0)2+p1×(μ-μ1)2
changing mu to p0μ0+p1μ1Substituting into the between-class variance formula to obtain the following between-class variance expression:
Vc＝2p0p1(μ1-μ0)2；
2.3) determining an image segmentation threshold:
using an optimization method, find the optimal threshold thr that maximizes the inter-class variance Vc*：
thr*＝max Vc；
2.4) using the optimal threshold value to segment the image to obtain a binary image BW of the mth frame*
The image segmentation result is shown in fig. 2(b) and fig. 4(b), where fig. 2(b) is a schematic diagram of the result of performing the object segmentation on the preprocessed image of the 15 th frame of the small target sequence in the scene with multiple strong noise points in the present example, and fig. 4(b) is a schematic diagram of the result of performing the object segmentation on the preprocessed image of the 15 th frame of the small target sequence in the ground background in the present example.
And step 3, establishing a pipeline.
Judging whether the current input frame is a first frame:
if the current input frame is a first frame, namely m is 1, the length of the pipeline is defined to be N, the pipe diameter is 2-3 times of the target size, the first frame binary image is copied to N parts to establish the pipeline, and the step 6 is skipped;
if the current input frame is not the first frame, i.e. the pipe has been established, step 4 is performed.
And 4, probabilistic pipeline filtering.
4.1) defining the pipeline parameters:
4.1a) defining the connected region of the N frames of targets in the pipeline and the centroid of the targets, and the formula is as follows:
maskk(i,j)＝{1,0}
Qk＝(xk,yk)
wherein, maskkRepresenting a connected region of a target in a k frame in a pipeline, namely a binary image with a background of 0 and a target of 1; qkRepresenting the coordinates of the mass center of the k frame target in the pipeline, wherein k is 1, 2.
4.1b) will target maskkThe result after morphological dilation treatment is used as the visual field bbox of the pipelinekThe definition is as follows:
bboxk＝dilate(maskk)
4.1c) according to maskk，QkAnd bboxkCalculating a feature matrix of the target, defined as follows:
wherein, αkRepresenting the shape feature vector of the k frame object in the pipeline, βkRepresenting a k-th frame of target space features in a pipelineEigenvectors, gammakFeature vector, w, representing the gray distribution of the k frame object in the pipelinek,hkRepresenting maskkWidth and height of (max)k,mink,meank,vark,sumkRespectively representing the maximum value, the minimum value, the mean value, the variance and the accumulated value of the target gray scale of the kth frame in the pipeline;
4.2) calculating the pipeline parameters:
4.2a) the pipeline receptive field:
mask＝∪bboxk；
4.2b) true target center position in the pipeline:
Q＝mean(Qk)
4.2c) feature vectors of real targets in the pipeline:
(α,β,γ)T＝mean(αk,βk,γk)T
wherein (·)TRepresenting a matrix transposition;
4.3) space constraint:
eliminating false targets which are not in the pipeline receptive field completely by using space constraint, calculating the intersection of the candidate targets and the pipeline receptive field by traversing all candidate targets of the input frame binary image BW, and judging the intersection:
if the intersection is not empty, the target is retained,
otherwise, the target is culled, i.e., only mask is retainedt∩ mask ≠ Φ candidates,
wherein the masktRepresenting a binary image of a tth candidate target of an input frame, wherein a mask represents a binary image receptive field of a real target accumulated in a pipeline, and phi represents that an intersection is empty;
4.4) further eliminating false targets by using gray distribution constraint:
calculating the gray distribution characteristic vector of the residual candidate target of the mth frame: gamma rayt＝(maxt,mint,meant,vart,sumt) Only the requirement of dist (gamma ) remainst)≤grayThe candidate targets for the condition(s) are,
wherein γ representsGray-scale distribution feature vector of real object, maxt,mint,meant,vart,sumtRespectively represent the maximum, minimum, mean, variance and cumulative sum of the t-th candidate target gray-scale value,graya threshold value representing a gray distribution constraint;
4.5) probabilistic target decision:
4.5a) calculating the confidence level between the remaining candidate targets in the input frame that meet the gray distribution constraint and the true target of the N-frame image in the pipeline:
wherein λ is1Representing the correlation coefficient, λ, between the true object and the candidate in terms of the shape characteristics of the object2Representing the correlation coefficient of the real target and the candidate target in terms of spatial features, dist (-) representing the distance between the two vectors, α and αtRepresenting the shape feature vector of the real object and the shape feature vector of the candidate object, β and β, respectivelytRespectively representing a real target space characteristic vector and a space characteristic vector of a candidate target;
4.5b) carrying out data normalization processing on the credibility scores of the candidate targets to obtain the probability that the candidate targets belong to the real targets:
qt＝softmax(pt)，
4.5c) output probability qtThe detection result of the largest candidate target as the real target of the mth frame is shown in fig. 2(c) and fig. 4(c), where fig. 2(c) is a diagram of the detection result of the example on the 15 th frame of the small target sequence in the multiple noisy scenes, and fig. 4(c) is a diagram of the detection result of the example on the 15 th frame of the ground background small target sequence.
And 5, updating the pipeline.
And updating the image to be processed in the pipeline by adopting a first-in first-out storage rule, namely moving out the first frame image in the pipeline, and moving the result of input frame target detection into the pipeline to the end.
And 6, detecting a task ending trigger.
And (5) making m equal to m +1, judging whether all the image sequences complete target detection:
if the input frame is not the last frame of the image sequence, namely M < M +1, returning to the step 1, and performing target detection on the next frame by using the updated pipeline;
if the input frame is the last frame of the image sequence, all frames of the image sequence have completed the target detection task, and the M-frame target detection results are output, as shown in fig. 3 and 5, where fig. 3 is a schematic diagram of the target detection results of all frames of the small target sequence in the multi-noisy-point scene in the present example, and fig. 5 is a schematic diagram of the target detection results of all frames of the small target sequence in the ground background in the present example.
The above description is only for the preferred embodiment of the present invention and does not limit the present invention in any way, and it is obvious that those skilled in the art can make any known variations on the main technical idea of the present invention, which fall within the technical scope of the present invention to be protected.
Claims (4)
1. A weak and small target detection method based on probabilistic pipeline filtering comprises the following steps:
(1) the method comprises the steps of inputting an M-th frame target image to be detected, wherein M is 1,2,., M, preprocessing the M-th frame target image by using a Robinson Guard spatial filter, and M is the total frame number of sequence images to be processed;
(2) performing target segmentation on the preprocessed input image by using a maximum inter-class variance method to obtain a binary image;
(3) building a pipeline: if the current input frame is a first frame, namely m is 1, defining the length of the pipeline as N and the pipe diameter as 2-3 times of the target size, copying N parts of binary images of the first frame to establish the pipeline, and jumping (6); otherwise, executing (4);
(4) and utilizing the established pipeline to perform target detection on the input image by using a probabilistic algorithm:
(4a) obtaining parameters of the size, the position and the characteristics of a corresponding target according to the existing N frames of images in the pipeline;
(4b) for the current input frame image, eliminating false targets which are not in the pipeline receptive field completely through space constraint;
(4c) and (3) further eliminating false targets by using gray distribution constraint, namely calculating gray distribution characteristic vectors of the residual candidate targets of the mth frame: gamma rayt＝(maxt,mint,meant,vart,sumt) T1, 2, T, only remaining to satisfy dist (γ, γ)t)≤grayConditional candidate objects, where γ represents the gray-scale distribution feature vector of the real object, T represents the number of remaining candidate objects in the mth frame, maxt,mint,meant,vart,sumtRespectively represent the maximum, minimum, mean, variance and cumulative sum of the t-th candidate target gray-scale value,graya threshold value representing a gray distribution constraint;
(4d) calculating the credibility between the residual candidate targets in the mth frame and the real targets of the N frames of images in the pipeline: p is a radical oft＝[λ1dist(α,αt)+λ2dist(β,βt)]-1Obtaining the credibility of a plurality of candidate targets of the mth frame belonging to the real target, and carrying out data normalization processing on the credibility scores to obtain the probability q of the candidate targets belonging to the real targett＝softmax(pt) Wherein λ is1Representing the correlation coefficient, λ, between the true object and the candidate in terms of the shape characteristics of the object2Representing the correlation coefficient of the real target and the candidate target in terms of spatial features, dist (-) representing the distance between the two vectors, α and αtRepresenting the shape feature vector of the real object and the shape feature vector of the candidate object, β and β, respectivelytRespectively representing a real target space characteristic vector and a space characteristic vector of a candidate target;
(4f) output probability qtThe largest candidate target is used as the real target of the mth frame;
(5) updating the pipeline: updating the image to be processed in the pipeline by adopting a first-in first-out storage rule, namely moving out the image of the first frame in the pipeline, and moving the detection result of the mth frame into the pipeline finally;
(6) let m be m + 1: if M < M +1, returning to (1); otherwise, ending the target detection task and outputting M frames of target detection results.
2. The method according to claim 1, wherein in (1) the object to be detected is preprocessed by using a Robinson Guard space filter, and the difference between the extreme values of the central pixel and the adjacent pixels is compared to realize background suppression and object enhancement so as to improve the signal-to-noise ratio of the input image; meanwhile, complete information of the weak and small targets is effectively reserved by setting a guard band based on template operation.
3. The method according to claim 1, wherein in (2), the target segmentation is performed on the preprocessed input image by using a maximum inter-class variance method, and the following is implemented:
(2a) obtaining the proportion of target and background pixels in the whole image: assuming that the division threshold value of the target and the background is thr, the pixels with the gray values larger than the division threshold value are the target, and the pixels with the gray values smaller than the division threshold value are the background, the proportion p of the target pixels in the whole image can be obtained1And target pixel mean μ1Background pixels in the proportion p of the whole image0And background pixel mean μ0And calculating to obtain the integral image mean value mu as p0μ0+p1μ1；
(2b) Calculating the inter-class variance Vc ═ p between the target and the background0×(μ-μ0)2+p1×(μ-μ1)2；
(2c) Determining an optimal threshold value of image segmentation: finding the optimal threshold thr*Making the inter-class variance Vc maximum;
(2d) and segmenting the image by using the optimal threshold value to obtain a binary image.
4. The method of claim 1, wherein: (4b) in the method, false targets which are not in the pipeline sensing field at all are removed through space constraint, namely, binary images are input into a probabilistic pipeline filter, suspicious targets which are not in the pipeline sensing field at all are removed, namely, masks are reservedt∩ mask ≠ Φ of candidate targets, where masktRepresenting the binary image of the tth candidate target, mask representing the pipelineAnd (4) internally accumulating the binary image receptive field of the real target, wherein phi represents that the intersection is empty.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010248459.9A CN111461011B (en) | 2020-04-01 | 2020-04-01 | Weak and small target detection method based on probabilistic pipeline filtering |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010248459.9A CN111461011B (en) | 2020-04-01 | 2020-04-01 | Weak and small target detection method based on probabilistic pipeline filtering |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111461011A true CN111461011A (en) | 2020-07-28 |
| CN111461011B CN111461011B (en) | 2023-03-24 |
Family
ID=71684343
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010248459.9A Active CN111461011B (en) | 2020-04-01 | 2020-04-01 | Weak and small target detection method based on probabilistic pipeline filtering |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111461011B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111983629A (en) * | 2020-08-14 | 2020-11-24 | 西安应用光学研究所 | Linear array signal target extraction device and extraction method |
| CN112016537A (en) * | 2020-10-27 | 2020-12-01 | 成都考拉悠然科技有限公司 | Comprehensive mouse detection method based on computer vision |
| CN112435249A (en) * | 2020-11-30 | 2021-03-02 | 天津津航技术物理研究所 | Dynamic small target detection method based on periodic scanning infrared search system |
| CN113379639A (en) * | 2021-06-25 | 2021-09-10 | 武汉大学 | Anti-interference detection tracking method for infrared target in complex environment |
| CN114693732A (en) * | 2022-03-07 | 2022-07-01 | 四川大学华西医院 | Weak and small target detection and tracking method |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102074022A (en) * | 2011-01-10 | 2011-05-25 | 南京理工大学 | Infrared image-based weak and small moving target detecting method |
| WO2016165064A1 (en) * | 2015-04-14 | 2016-10-20 | 中国科学院自动化研究所 | Robust foreground detection method based on multi-view learning |
| CN106469313A (en) * | 2016-09-30 | 2017-03-01 | 中国科学院光电技术研究所 | A kind of detection method of small target of caliber adaptive space-time domain filtering |
| CN109003277A (en) * | 2017-06-07 | 2018-12-14 | 中国航空工业集团公司洛阳电光设备研究所 | A kind of infrared small target in complex background detection method and device |
-
2020
- 2020-04-01 CN CN202010248459.9A patent/CN111461011B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102074022A (en) * | 2011-01-10 | 2011-05-25 | 南京理工大学 | Infrared image-based weak and small moving target detecting method |
| WO2016165064A1 (en) * | 2015-04-14 | 2016-10-20 | 中国科学院自动化研究所 | Robust foreground detection method based on multi-view learning |
| CN106469313A (en) * | 2016-09-30 | 2017-03-01 | 中国科学院光电技术研究所 | A kind of detection method of small target of caliber adaptive space-time domain filtering |
| CN109003277A (en) * | 2017-06-07 | 2018-12-14 | 中国航空工业集团公司洛阳电光设备研究所 | A kind of infrared small target in complex background detection method and device |
Non-Patent Citations (2)
| Title |
|---|
| 丁云等: "天空背景下红外弱小目标检测算法研究", 《现代电子技术》 * |
| 闻凯: "基于复杂融合特征与灰度-纹理直方图描述子的红外弱小目标检测追踪算法", 《科学技术与工程》 * |
Cited By (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111983629A (en) * | 2020-08-14 | 2020-11-24 | 西安应用光学研究所 | Linear array signal target extraction device and extraction method |
| CN111983629B (en) * | 2020-08-14 | 2024-03-26 | 西安应用光学研究所 | Linear array signal target extraction device and extraction method |
| CN112016537A (en) * | 2020-10-27 | 2020-12-01 | 成都考拉悠然科技有限公司 | Comprehensive mouse detection method based on computer vision |
| CN112016537B (en) * | 2020-10-27 | 2021-01-08 | 成都考拉悠然科技有限公司 | Comprehensive mouse detection method based on computer vision |
| CN112435249A (en) * | 2020-11-30 | 2021-03-02 | 天津津航技术物理研究所 | Dynamic small target detection method based on periodic scanning infrared search system |
| CN112435249B (en) * | 2020-11-30 | 2024-04-16 | 天津津航技术物理研究所 | Dynamic small target detection method based on circumferential scanning infrared search system |
| CN113379639A (en) * | 2021-06-25 | 2021-09-10 | 武汉大学 | Anti-interference detection tracking method for infrared target in complex environment |
| CN113379639B (en) * | 2021-06-25 | 2022-08-05 | 武汉大学 | Anti-interference detection tracking method for infrared target in complex environment |
| CN114693732A (en) * | 2022-03-07 | 2022-07-01 | 四川大学华西医院 | Weak and small target detection and tracking method |
| CN114693732B (en) * | 2022-03-07 | 2022-11-25 | 四川大学华西医院 | Weak and small target detection and tracking method |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111461011B (en) | 2023-03-24 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111461011B (en) | Weak and small target detection method based on probabilistic pipeline filtering | |
| CN107872644B (en) | Video monitoring method and device | |
| CN112819772B (en) | High-precision rapid pattern detection and recognition method | |
| CN109658424B (en) | Improved robust two-dimensional OTSU threshold image segmentation method | |
| US8094936B2 (en) | Method and apparatus to segment motion area in real-time to detect motion in surveillance camera system | |
| CN110909591B (en) | Self-adaptive non-maximum suppression processing method for pedestrian image detection by using coding vector | |
| CN111723644A (en) | Method and system for detecting occlusion of surveillance video | |
| CN113537099B (en) | Dynamic detection method for fire smoke in highway tunnel | |
| CN110400294B (en) | Infrared target detection system and detection method | |
| CN110580709A (en) | Target detection method based on ViBe and three-frame differential fusion | |
| CN113888461A (en) | Method, system and equipment for detecting defects of hardware parts based on deep learning | |
| CN114897728A (en) | Image enhancement method and device, terminal equipment and storage medium | |
| CN110827262A (en) | Weak and small target detection method based on continuous limited frame infrared image | |
| CN111563849A (en) | Observation image denoising method and system | |
| CN115393734A (en) | SAR image ship contour extraction method based on fast R-CNN and CV model combined method | |
| CN115359258A (en) | Weak and small target detection method and system for component uncertainty measurement | |
| CN109308709B (en) | Vibe moving target detection algorithm based on image segmentation | |
| CN110910428B (en) | Real-time multi-target tracking method based on neural network | |
| CN113592740A (en) | Image noise removing method in air tightness detection based on artificial intelligence | |
| CN110502968B (en) | Method for detecting infrared small and weak moving target based on track point space-time consistency | |
| CN115511804A (en) | Offshore infrared dim target detection method based on pipeline adaptive filtering | |
| CN111881837A (en) | Video SAR moving target detection method based on shadow extraction | |
| CN113724290B (en) | Multi-level template self-adaptive matching target tracking method for infrared image | |
| CN108038872B (en) | Dynamic and static target detection and real-time compressed sensing tracking research method | |
| CN114155425B (en) | Weak and small target detection method based on Gaussian Markov random field motion direction estimation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |