CN109215018A - Based on Canny operator and the morphologic ship detecting method of Gauss - Google Patents
Based on Canny operator and the morphologic ship detecting method of Gauss Download PDFInfo
- Publication number
- CN109215018A CN109215018A CN201810967927.0A CN201810967927A CN109215018A CN 109215018 A CN109215018 A CN 109215018A CN 201810967927 A CN201810967927 A CN 201810967927A CN 109215018 A CN109215018 A CN 109215018A
- Authority
- CN
- China
- Prior art keywords
- ship
- pixel
- gauss
- profile
- canny operator
- 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
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/0002—Inspection of images, e.g. flaw detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/13—Edge detection
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/136—Segmentation; Edge detection involving thresholding
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06T7/00—Image analysis
- G06T7/10—Segmentation; Edge detection
- G06T7/155—Segmentation; Edge detection involving morphological operators
Landscapes
- Engineering & Computer Science (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Quality & Reliability (AREA)
- Image Analysis (AREA)
Abstract
The invention proposes one kind to be based on Canny operator and the morphologic ship detecting method of Gauss, and this method carries out ship detecting from the picture frame of maritime affairs monitor video, obtains ship profile information, and identify the ship of different imaging sizes.This method includes ship contours extract, and noise is eliminated and three parts of ship contour reconstruction.First with all possible ship profile information in Canny operator extraction picture frame；Denoising, the corresponding contour pixel of removal noise, such as background image information are further smoothed to the ship profile expansion of Canny operator extraction followed by gaussian filtering operator；Finally using the opening operation operation reconstruct ship profile of morphological method, final ship detecting result is obtained.The method is applied to the ship detecting in different traffic and ship wake interference scene, verifies detection performance of the invention.Method of the invention obtains higher accuracy rate in different test scenes, has obtained preferable detection effect.
Description
Technical field
The present invention relates to waterborne target detection field, in particular to the ship detecting based on visible light technology.
Background technique
Existing waterborne target detection algorithm can be roughly divided into following three classes: the waterborne target detection based on infrared technique,
Waterborne target detection based on Radar Technology and the waterborne target detection based on visible light technology.
There is higher gray value than background based on the weak signal target of infrared image so that naval target in infrared image with
Background difference is significant, this advantage makes waterborne target detection side of numerous experts and scholars all in actively research infrared image
Method.The sky-line in infrared image has more specific structure feature, can efficiently differentiate the water surface and the sky areas of image.
For this feature in infrared image, Wang Yuehang et al. is detected and is determined to the sea horizon in infrared image, and is utilized
The mean-shift algorithm of weighting detects infrared image target, and Zheng Hua also utilizes the water day detected in infrared image
Line determines the target zone of infrared image.
In the casualty region far from land, the distress personnel that infrared image acquiring equipment takes is usual in the picture
Small object is shown as, and the imaging region of distress personnel occurs frequently near the sky-line.Therefore, many scholars determine first
The specific location of image Small Target is found in sky-line region in infrared image according to the sky-line position detected, for sea
Upper search and rescue provide visual information foundation.But infrared detection system has noise bigger than normal, between target and background compared with edge
Fuzzy feature, and the infrared image of boat-carrying video camera shooting is often shaken, this causes larger interference to naval target detection.
Waterborne target detection effect based on Radar Technology is affected by sea clutter, the sea clutter in radar image
Non-Gaussian feature and non-stationary statistical characteristic make the detection effect of the waterborne target based on radar by more acute challenge.
Therefore, research is unfolded to the mechanism of production of sea clutter in many scholars, constructs corresponding sea clutter distributed model.Researcher's discovery
Wei Buer is distributed (Weibull distribution), rayleigh distributed (Rayleigh Distribution), K distribution and logarithm
Four kinds of models of normal distribution can preferably be fitted sea clutter distribution.Radar image can be effectively suppressed in method based on Power estimation
Sea clutter.On this basis, all kinds of water surface mesh are extracted using the methods of fractal theory, empirical mode decomposition and target's feature-extraction
Mark, such as the fast Small object of the water surface low (height), surface vessel target etc..
The Weigh sensor degree of ship is higher and higher, and the waterborne target detection research based on machine vision method causes
More and more concerns.Since the variation of the factors such as weather, illumination causes scene dynamics waterborne to change, so that the water based on visible light
Area Objects detection method faces great challenge.Ren Lei, Ran Xin et al. expand multinomial for the target detection waterborne of visible light
Research, and point out that target detection correlative study waterborne is unfolded using visible light technology needs to overcome the difficulty in terms of following four:
1. pixel shared by target is less, background and noise occupy most of pixels of visible light video sequence, and signal noise ratio (snr) of image is smaller；
2. the different sea situations such as the irradiation of the heave of waves, sunlight and sleet mist and weather condition bring larger interference to waterborne target detection；
3. target waterborne may be blocked briefly by wave, cause detection algorithm in the visible light video image comprising target waterborne
Waterborne target is lost, the detection accuracy of algorithm is reduced；4. the picture pick-up device of acquisition waterborne target detection video is often fixed to ship
In oceangoing ship or helicopter.Picture pick-up device is subjected to wind, wave, gushes, the influence of the extraneous factors such as ship or helicopter, may cause
There are jitter phenomenons for the video pictures of picture pick-up device acquisition, this requires waterborne target detection algorithm to have preferable robustness.
Summary of the invention
It is a kind of based on Canny operator and the morphologic ship detecting method of Gauss the purpose of the invention is to provide, it is right
The ship of normal imaging size carries out target detection in video image frame, identifies ship target.Normal imaging size in the present invention
Ship image refer to: ship imaging size is not less than the 0.15% of the frame image actual size, and re-imaging length or width are not
Less than 13 pixels.This method includes ship contours extract, and noise is eliminated and three parts of ship contour reconstruction.By by this ship
The ship detecting that oceangoing ship detection method is applied to different water transportation states and ship wake interferes, verification result show the present invention
Method higher accurate rate is obtained to the identification of ship profile in different test scenes.
In order to achieve the above objectives, the present invention proposes that one kind is based on Canny operator and the morphologic ship detecting method of Gauss,
Suitable for the picture frame to video, the detection of the ship image border of normal imaging size, comprising the following steps:
S1, picture noise is removed using 2-d gaussian filters device to original maritime affairs monitor video picture frame, is denoised
Ship picture frame；
S2, using Canny operator, extract the ship profile information in the ship picture frame of the denoising；
S3, according to the ship profile information, calculate the weighting parameter of adaptive Gaussian mixture model device, joined according to the weight
Number, using adaptive Gaussian mixture model device to the ship profile information, further smooths denoising；
S4, the processing result of step S3 is obtained final using the opening operation operation reconstruct ship profile of morphological method
Ship detecting result.
The step S1 the following steps are included:
S11, according to the kernel function of the 2-d gaussian filters deviceObtain Gaussian convolution square
Battle array, the element of the Gaussian convolution matrix is the weight of image convolution；Wherein w is the distance between pixel and x axis, and h generation
For the table pixel to the distance of y-axis, σ is the standard deviation of gaussian kernel function；
S12, to any pixel of the original maritime affairs monitor video picture frame, by the neighborhood territory pixel of the pixel and the height
This convolution matrix carries out convolution operation, and the weighted average of pixel in neighborhood is taken to the pixel of each pixel, obtains the pixel
Pixel value after point Gauss denoising, realizes and denoises to picture frame, obtain the denoising ship picture frame.
The Gaussian convolution matrix is the matrix of 3 × 3 sizes.
The step S2 the following steps are included:
S21, the gradient direction and gradient amplitude for calculating all pixels in the ship picture frame of the denoising extract ship figure
As all probable edge information of frame；
S22, the non-maximum value suppression mechanism using gradient filter out the profile letter of the non-ship in the probable edge information
Breath rejects the ship marginal point of mistake, obtains true ship edge pixel；
S23, ship marginal point false in the ship edge pixel is further excluded using dual-threshold voltage, extraction obtains
The true ship profile information.
The step S21 the following steps are included:
S211, the convolution mask matrix that x-axis is setThe convolution mask matrix of y-axis
For each pixel in the ship picture frame of the denoising, C is utilized in the direction x and y respectivelyx、Cy, to the gray scale of pixel
Value carries out convolution, obtains gradient information, convolution method are as follows:
Wherein, I (i, j) is the gray value of pixel (i, j), and P (i, j) and Q (i, j) respectively represent the pixel in x and y
The gradient information in direction；
S212, the gradient information according to the pixel (i, j) obtained in step S211 in the direction x and y, calculate the pixel
Gradient amplitude G (i, j) and gradient direction θ (i, j):
G (i, j) is greater than the pixel collection of preset threshold as all probable edge information of ship picture frame.
The step S23 the following steps are included:
S231, strong edge threshold value T1 and weak edge threshold T2 is set, wherein T1 > T2；
S232, G (i, j) indicate the gradient amplitude of pixel (i, j), when G (i, j) is greater than T1, then make pixel (i, j)
For strong edge point, all strong edge pixels are all ship profile points；If T1 > G (i, j) > T2, by the pixel (i, j)
As weak edge pixel point；
S233, strong edge point and weak marginal point are then connected according to eight connectivity metric, obtain the final ship wheel of Canny operator
Wide information.
In the step S3, comprising steps of
Ship profile in S31, the ship profile information for enabling step S2 extract is S (x, y), and β is adaptive Gaussian mixture model
The weighting parameter of device, E (β) indicate the desired value of the adaptive Gaussian mixture model device, and G (x, y) indicates the adaptive Gauss filter
Wave device operator, operator * indicate the convolution operation of G (x, y) and S (x, y), and ▽ operator is derivative operation symbol, and parameter lambda is characterization E
The coefficient of (β) convergence rate, the then expression formula of E (β) are as follows:
E (β)=∫ ∫ [(S (x, y)-G (x, y) * S (x, y))2+λ((▽β)-1)2],
Acquire the minimum value of E (β)；
S32, according to the E (β) minimum value, find out the value of β, as the adaptive Gaussian mixture model device best initial weights join
Number；Enable Sp(x, y) is the filtered ship profile of adaptive Gaussian mixture model device, and the sef-adapting filter is utilized according to the value of β
FormulaFiltering and noise reduction is further smoothed to ship profile S (x, y).
The step S4 the following steps are included:
S41, etching operation is carried out to the ship profile information for further smoothing denoising obtained in step S3, completely
Remove the edge of background object；
S42, expansive working is carried out to the ship profile information after etching operation, reconstructs true ship profile, obtained most
Whole ship detecting result.
Compared with prior art, the target inspection of a kind of fusion Canny operator and Gauss morphology operations provided by the invention
Survey method tentatively obtains image sequence first with Canny operator to realize the ship target detection of normal imaging size
Ship profile.On this basis, noise in the ship profile of Canny operator extraction is filtered out using gaussian filtering operator to correspond to
Contour pixel.Finally, operating reconstruct ship profile using morphologic opening operation, final ship detecting result is obtained.Phase
It is answering the experimental results showed that, under different detection environment, the ship target detection method of the normal imaging size of proposition
Obtain false detection rate more lower than traditional ship detecting method and omission factor.
Detailed description of the invention
In order to illustrate more clearly of technical solution of the present invention, attached drawing needed in description will be made simply below
It introduces, it should be apparent that, the accompanying drawings in the following description is one embodiment of the present of invention, and those of ordinary skill in the art are come
It says, without creative efforts, is also possible to obtain other drawings based on these drawings:
Fig. 1 is the flow chart of the invention based on Canny operator and the morphologic ship detecting method of Gauss.
Fig. 2 is the gradient calculation schematic diagram of the pixel based on NMS mechanism.
Fig. 3 is the not busy scene figure of traffic in video frame of the invention.
Fig. 4 is the scene figure of heavy traffic in video frame of the invention.
Fig. 5 is the scene figure of ship's navigation tail interference in video frame of the invention.
Fig. 6 is the testing result that method of the invention is used to Fig. 3.
Fig. 7 is the testing result that normal Gaussian algorithm is used to Fig. 3.
Fig. 8 is the testing result that method of the invention is used to Fig. 4.
Fig. 9 is the testing result that normal Gaussian algorithm is used to Fig. 4.
Figure 10 is the testing result that method of the invention is used to Fig. 5.
Figure 11 is the testing result that normal Gaussian algorithm is used to Fig. 5.
Specific embodiment
The technical features, objects and effects for a better understanding of the present invention with reference to the accompanying drawing carry out more the present invention
To describe in detail.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, it is not used to limit this
Patent of invention.It should be noted that being all made of very simplified form in these attached drawings and using non-accurate ratio, only use
In convenience, clearly aid in illustrating the invention patent.
The present invention proposes a kind of based on Canny operator and Gauss morphology (Canny operator and Gaussian-
Morphology, CGM) ship detecting method, suitable for the picture frame to video, the ship image of normal imaging size
Detection.The ship image of normal imaging size refers in the present invention: ship imaging size is not less than the frame image actual size
0.15%, and re-imaging length or width are not less than 13 pixels.Ship monitor source video sequence of the invention is in Shanghai marine board
CCTV (the electronics cruise closed-circuit television Closed-Circuit of (Maritime Safety Administration, MSA)
Television System) system.
Ship profile information is to discriminate between the important feature of ship and its ambient enviroment, ship detecting method of the present invention
It is to be realized by extracting the ship profile information in image.
As shown in Figure 1, detection method detailed process according to the present invention are as follows:
S1, picture noise is removed using 2-d gaussian filters device to original maritime affairs monitor video picture frame, is denoised
Ship picture frame；
S2, Canny operator, the extraction ship profile information denoised in ship picture frame are utilized；
S3, according to the ship profile information, calculate the weighting parameter of adaptive Gaussian mixture model device, joined according to the weight
Number, using adaptive Gaussian mixture model device to the ship profile information, further smooths denoising；
S4, the processing result of step S3 is obtained final using the opening operation operation reconstruct ship profile of morphological method
Ship detecting result.
Step S1 is specific as follows:
Since CCTV system is highly susceptible to the external interferences such as picture pick-up device shake during acquiring ship monitoring image,
This causes video easily by noise pollution, needs the picture frame denoising first to ship monitor video, excludes ship video figure
The potential noise as in.It can be fitted picture noise well in view of Gaussian Profile, we are removed using 2-d gaussian filters device
Picture noise.The kernel function of the 2-d gaussian filters device is as follows:
Wherein w is the distance between pixel and x-axis, and h represents the pixel to the distance of y-axis, and σ is gaussian kernel function
Standard deviation；
Gaussian convolution matrix is obtained by gaussian kernel function, the element of the matrix is the weight of image convolution.For giving
Any pixel of fixed ship picture frame, by the neighborhood territory pixel of the pixel and the convolution operation of convolution matrix, to each pixel
The pixel of point takes the weighted average of pixel in neighborhood, the pixel value after obtaining pixel Gauss denoising.
The original pixels gray value of noise is higher than its adjacent pixel, and above-mentioned Gaussian smoothing operation can obviously reduce the ash of noise
Angle value.The order of convolution matrix is higher, and the denoising effect of image is better.However, the convolution matrix of high-order can reduce the side of image
Edge detection performance.Existing research shows that the convolution matrix of 3 × 3 ranks is classical convolution matrix, there is preferable image denoising
Effect, and edge detection demand is taken into account, therefore use 3 × 3 rank matrixes as the Gaussian convolution matrix of default in the present invention.
In the more serious image sequence of noise interference, three basic demands that outstanding detective operators should have are 1.
Detective operators can be found that edge as much as possible；2. the edge detected and actual edge are coincide；3. actual edge should only by
Mark is primary, and noise should not be marked as edge.Canny operator meets above-mentioned requirements, and to the detection effect of object edge
Better than edge detection operators such as Prewitt, Sobel and Laplacian.Therefore the present invention uses Canny first in step s 2
Operator realizes the preliminary extraction to ship profile.It is specific as follows:
S21, to the ship picture frame after step S1 noise reduction, calculate the gradient direction and gradient amplitude of all pixels, extract
All marginal informations of ship picture frame；
The convolution mask matrix of x-axis is set
The convolution mask matrix of y-axis
For each pixel in the denoising ship picture frame, C is utilized in the direction x and y respectivelyx、 Cy, to pixel
Gray value carry out convolution, obtain gradient information, convolution method are as follows:
Wherein, I (i, j) is the gray value of pixel (i, j), and P (i, j) and 22Q (i, j) respectively represent the pixel in x
With the gradient information in the direction y；
G (i, j) is enabled to indicate the gradient amplitude of pixel (i, j), θ (i, j) indicates the gradient direction of pixel (i, j),
Obtain the gradient amplitude and gradient direction of the pixel；G (i, j) is bigger, and (i, j) is that the probability of marginal point is higher.
Therefore, G (i, j) is greater than the pixel collection of preset threshold as all probable edge information of ship picture frame here.
S22, NMS (the non-maximum value of Non-maximum suppression inhibits) mechanism using gradient, filter out step
The profile information for the non-ship that S21 is obtained excludes false ship marginal point, finds true ship edge pixel；
As shown in Fig. 2, put pixel centered on enabling Pc, P1, P2, the neighborhood point that P3 and P4 are Pc, endpoint g1 and g2 in figure
Line is the gradient direction of just point Pc.
The judgment mechanism of NMS is, if the gradient amplitude of central point Pc is less than the gradient amplitude of g1 (or g2), NMS
Mechanism thinks that the point Pc of the picture frame is not belonging to the profile point of ship.
S23, false ship marginal point is further excluded using dual-threshold voltage, extract true ship profile.
The principle of dual-threshold voltage are as follows: set strong edge threshold value T1 and weak edge threshold T2 first, wherein T1 > T2；G(i,j)
It indicates the gradient amplitude of pixel (i, j), when G (i, j) is greater than T1, then pixel (i, j) is used as strong edge point, Suo Youqiang
Edge pixel point is all ship profile point；If T1 > G (i, j) > T2, it regard the pixel (i, j) as weak edge pixel point
(T2 can be used as above-mentioned preset threshold here, it would be possible that marginal information includes the pixel that all G (i, j) are greater than T2)；
Strong edge point and weak marginal point are then connected according to eight connectivity metric, obtains the final ship profile letter of Canny operator
Breath.Eight connectivity metric then refers to, if having non-zero pixels (i ', j ') to exist in 8 adjacent domains of pixel (i, j), then to pixel
Point (i, j) and (i ', j ') line are as contour line.So far, the edge detection of canny operator is completed.
Step S3 is specific as follows:
Enable the ship profile in the ship profile information that S (x, y) is step S2 extraction；Sp(x, y) is adaptive Gauss filter
The filtered ship profile of wave device；Ship profile S after known adaptive Gaussian mixture modelp(x, y) is approximately equal to S (x, y) and S
The sum of (x, y) second dervative, as shown by the equation,
β is the weighting parameter of adaptive Gaussian mixture model device, determines adaptive Gaussian mixture model device to S (x, y) edge pixel
Blur effect, β2It is the variance of the adaptive Gaussian mixture model device.The optimal effectiveness of adaptive Gaussian filter is obtained, first
Find optimal β value.
E (β) is enabled to indicate the desired value of the adaptive Gaussian mixture model device, G (x, y) indicates the adaptive Gaussian mixture model device
Operator, operator * indicate the convolution operation of G (x, y) and S (x, y), and ▽ operator is derivative operation symbol, and parameter lambda is that characterization E (β) is received
The coefficient of speed is held back, then the expression formula of E (β) are as follows:
E (β)=∫ ∫ [(S (x, y)-G (x, y) * S (x, y))2+λ((▽β)-1)2] (9)
Acquire the minimum value of E (β)；According to E (β) minimum value, β value is found out；Using β as the optimal of adaptive Gaussian mixture model device
Weighting parameter.The sef-adapting filter further smooths ship profile S (x, y) using formula (8) according to the value of β
Filtering and noise reduction.
Step S4 is specific as follows:
Since the Canny operator in step S2 also remains the corresponding profile of part background of image.For example, common is floating
Mark.Since buoy is not picture noise, Canny operator can not eliminate the corresponding marginal information of buoy.Utilize buoy etc.
The imaging size of background object is much smaller than the imaging size of ship, and the present invention is further gone using the etching operation of Morphology Algorithm
Except the background edge of the ship profile of Canny operator extraction.
S41, etching operation is carried out to the ship profile information of the smoothing denoising processing obtained in step S3, completely removes back
The edge of scenery body；
S42, expansive working is carried out to the ship profile information after etching operation, reconstructs true ship profile, obtained most
Whole ship detecting result.
Test experience and analysis are carried out to method of the invention below.
Test experience of the invention uses 10 operating system of Win based on 64, and the dominant frequency of CPU is 3.4GHz, memory
For 8GB, the platform of emulation experiment is Python (2.7 version) and the library OpenCV (2.4.13 version).
Maritime affairs monitor video of the invention uses PORT OF SHANGHAI CCTV monitor video, and the huge container throughput in PORT OF SHANGHAI is led
Causing the navigation channel in the innerland of PORT OF SHANGHAI is domestic inland water transport one of navigation channel the busiest.Therefore ship is carried out using its monitor video
Oceangoing ship detection has certain practical significance.
Contrasting detection method of the invention is traditional gaussian filtering ship detecting algorithm.Although frame difference method is background modeling
One of with object detection field common method, but frame difference method is not appropriate for ship detecting of the invention.This is primarily due to this
The ship of the ship detecting video of invention, parts of images frame is in state of casting anchor, and frame difference method can regard as the ship to cast anchor
Background.Therefore, ship monitor video of the invention is detected using frame difference method, detection accuracy is lower.Another is often
Detection method is the ship detecting based on Gaussian filter, since Gaussian filter is that the intensity profile based on image is distinguished
Anchored vessel will not be labeled as background by ship and background pixel, therefore, the filter based on Gauss method.Based on this, Wo Menfen
Not Ying Yong this chapter propose method and traditional Gauss method the ship in maritime affairs monitor video is detected.For convenience,
The ship detecting algorithm and traditional gaussian filtering ship detecting algorithm that this chapter is proposed are denoted as CGM and Gaussian respectively.
The CCTV video of acquisition is divided into not busy traffic, heavy traffic, ship's navigation tail three allusion quotations of interference by us
Type scene difference is as shown in figure 3, figure 4 and figure 5.Heavy traffic and the not busy scene of traffic are used to assess under different traffic,
The detection performance of the method for the present invention, ship's navigation tail interference scene are used to verify the robustness of the method for the present invention.This experiment is fixed
The scene of adopted heavy traffic is that the ships quantity that picture frame includes is no less than 10, and the not busy scene of traffic refers to picture frame
Ships quantity be not more than 5.Ship detecting is carried out with CGM and Gaussian detection algorithm respectively to three kinds of scenes.Each field
The video frame rate of scape is that 30 frames are (fps) per second, and the resolution ratio of every frame is 1280 × 720.The not busy scene of traffic shares 539 frames
Picture, heavy traffic scene share 660 frame pictures, and the corresponding video length of ship's navigation tail interference scene is 330 frames.Due to
The maritime affairs monitor video that we acquire, ship are mobile slow.If ship does not have apparent operational configuration variation (if turned, to hand over
Fork can be met, accelerate etc.), the difference between video frame can be ignored, therefore we use the key frame of ship monitor video
Carry out the detection performance of evaluation algorithms.By the detection for extracting 100 key frames in the ship monitor video under three kinds of scenes respectively
As a result, the performance of two kinds of ship detecting algorithms of statistics and comparison.
The evaluation index of the ship detecting method of the method for the present invention is recall rate and accurate rate.Recall rate (Re) and accurate
Shown in the definition such as formula (10) of rate (Pr) and (11).Re shows to have been detected by ship target, and target is true ship
Accuracy.Re value is higher, and expression testing result is better.Parameter Pr represents the accuracy of detector.Pr value is higher to be indicated to be detected
The quantity of operator erroneous detection is fewer.
Wherein T indicates the ships quantity that algorithm is correctly detecting, TFIndicate the ships quantity of algorithm missing inspection, FTIndicate algorithm
The ships quantity of erroneous detection.
Detection effect is as described below:
Fig. 6, Fig. 7 are respectively CGM algorithm and Gaussian algorithm to not busy the 2nd frame image of scene of traffic shown in Fig. 3
Testing result.As shown in Figure 6, Figure 7, the CGM algorithm of proposition can almost detect all ships of the frame, and Gaussian algorithm
Testing result show that the algorithm detects multiple false-alarm targets.Fig. 6, Fig. 7 are shown in the not busy situation of maritime traffic,
The detection effect of CGM algorithm is significantly better than the testing result of Gaussian algorithm.
Fig. 8, Fig. 9 are respectively the inspection of CGM algorithm and Gaussian algorithm to the 14th frame image of heavy traffic scene shown in Fig. 4
Survey result.As shown in Figure 8, Figure 9, at sea in the case where heavy traffic, the detection performance of CGM algorithm is held essentially constant, and
The false detection rate and omission factor situation of Gaussian algorithm obviously increase.
The traffic of table 1 is not busy and heavy traffic scene in, the recall rate of two kinds of algorithms and accurate rate distribution also demonstrate
Above-mentioned analysis.For scene not busy for traffic, the recall rate of CGM method is 0.96, and the recall rate of Gaussian operator
The recall rate of the recall rate ratio Gaussian operator of only 0.70, CGM algorithm is high by 37%.The recall rate index of CGM algorithm shows
CGM can probably detect in monitor video 96% ship, and Gaussian operator recall rate shows that it can about detect sea
70% ship in thing video sequence.But, CGM algorithm is identical with the detection accuracy of Gaussian algorithm, the inspection of two kinds of algorithms
Surveying precision is 94%.For heavy traffic scene, the Re of Gauss operator is 0.47, approximately equal to CGM operator recall rate
Half.Gauss operator recall rate index explanation, the algorithm cannot effectively cope with the ship detecting of maritime traffic busy state
Task.It is only 0.78 that Gauss operator, which detects accurate rate, this has also confirmed above-mentioned analysis.And propose CGM algorithm recall rate and
Accurate rate is respectively 0.96 and 0.91, this illustrates also obtain preferable ship inspection in the case of the heavy traffic at sea of CGM algorithm
Survey effect.Therefore, one can consider that the CGM algorithm proposed can obtain preferable inspection under heavy traffic or not busy state
Survey effect.
In order to further verify the robustness of CGM algorithm, to shown in fig. 5, contain ship's navigation tail interference scene
Maritime affairs monitor video carries out ship detecting experiment.Figure 10, Figure 11 are respectively CGM algorithm and Gaussian algorithm, are navigated to shown in Fig. 5
The testing result of the 3rd frame image of end of line mark interference scene.As shown in Figure 10, Figure 11, it is produced during white ship's navigation bright
Aobvious navigation tail, the navigation tail have greatly challenged the robustness of ship detecting algorithm.Figure 10 shows the navigation tail of ship
There is no reduction CGM algorithm detection performance, that is, the CGM method proposed will not navigate by water tail and be identified as ship mark, illustrate CGM
Algorithm can preferably inhibit ship's navigation tail.And Figure 11 is shown, the detection performance of Gaussian operator receives ship boat
The navigation tail of ship is identified as the ship in movement by the severe jamming of end of line mark, i.e. Gaussian operator.
Table 1 shows that the recall rate and accurate rate of the corresponding CGM algorithm of navigation tail interference scene are respectively 0.90 He
0.98.Recall rate and accurate rate equal held stationary of the CGM algorithm under three kinds of different scenes, this illustrates CGM algorithm not shipmate
Oceangoing ship, which detects, has preferable robustness under scene.The ship detecting for navigating by water the corresponding Gaussian algorithm of tail interference scene is average
Accurate rate and average recall rate are respectively 0.88 and 0.60, illustrate that Gaussian algorithm average detected precision is about 88%, and high
This operator is only able to detect in video sequence about 60% ship, i.e. Gauss operator is high to the average omission factor of maritime affairs monitor video
Up to 40%.
According to table 1, traffic is busy, heavy traffic, ship's navigation tail interfere in three typical scenes, CGM operator
Ship detecting performance be better than Gaussian operator, average accuracy and evaluation recall rate be 0.94.Based on above-mentioned quantitative
With qualitative analysis it is found that under different maritime affairs monitoring scenes, CGM algorithm can provide stable ship detecting performance.
The detection performance of CGM algorithm and Gaussian method under 1 different scenes of table
The above description is merely a specific embodiment, but scope of protection of the present invention is not limited thereto, any
Those familiar with the art in the technical scope disclosed by the present invention, can readily occur in various equivalent modifications or replace
It changes, these modifications or substitutions should be covered by the protection scope of the present invention.Therefore, protection scope of the present invention should be with right
It is required that protection scope subject to.
Claims (9)
1. one kind is based on Canny operator and the morphologic ship detecting method of Gauss, suitable for the image to maritime affairs monitor video
In frame, the detection of the ship image border of normal imaging size, which comprises the following steps:
S1, picture noise, the ship denoised are removed using 2-d gaussian filters device to original maritime affairs monitor video picture frame
Picture frame；
S2, using Canny operator, extract the ship profile information in the ship picture frame of the denoising；
S3, according to the ship profile information, calculate the weighting parameter of adaptive Gaussian mixture model device, according to the weighting parameter,
Using adaptive Gaussian mixture model device to the ship profile information, denoising is further smoothed；
S4, final ship is obtained using the opening operation operation reconstruct ship profile of morphological method to the processing result of step S3
Oceangoing ship testing result.
2. being based on Canny operator and the morphologic ship detecting method of Gauss as described in claim 1, which is characterized in that institute
The ship image for stating normal imaging size refers to: 0.15% of ship imaging size not less than the frame image actual size, and at
As length or width is not less than 13 pixels.
3. being based on Canny operator and the morphologic ship detecting method of Gauss as described in claim 1, which is characterized in that institute
State step S1 the following steps are included:
S11, according to the kernel function of the 2-d gaussian filters deviceObtain Gaussian convolution matrix, institute
The element for stating Gaussian convolution matrix is the weight of image convolution；Wherein w is the distance between pixel and x-axis, and h represents the picture
For vegetarian refreshments to the distance of y-axis, σ is the standard deviation of gaussian kernel function；
S12, to any pixel of the original maritime affairs monitor video picture frame, the neighborhood territory pixel of the pixel and the Gauss are rolled up
Product matrix carries out convolution operation, and the weighted average of pixel in neighborhood is taken to the pixel of each pixel, obtains pixel height
Pixel value after this denoising, realizes and denoises to picture frame, obtain the ship picture frame of the denoising.
4. being based on Canny operator and the morphologic ship detecting method of Gauss as claimed in claim 3, which is characterized in that institute
State the matrix that Gaussian convolution matrix is 3 × 3 sizes.
5. as claimed in claim 1 or 3 be based on Canny operator and the morphologic ship detecting method of Gauss, which is characterized in that
The step S2 the following steps are included:
S21, the gradient direction and gradient amplitude for calculating all pixels in the ship picture frame of the denoising extract ship picture frame
All probable edge information；
S22, the non-maximum value suppression mechanism using gradient, filter out the profile information of the non-ship in the probable edge information,
The ship marginal point for rejecting mistake, obtains true ship edge pixel；
S23, ship marginal point false in the ship edge pixel is further excluded using dual-threshold voltage, extraction obtains true
The ship profile information.
6. being based on Canny operator and the morphologic ship detecting method of Gauss as claimed in claim 5, which is characterized in that institute
State step S21 the following steps are included:
S211, the convolution mask matrix that x-axis is setThe convolution mask matrix of y-axis
For each pixel in the ship picture frame of the denoising, C is utilized in the direction x and y respectivelyx、Cy, to the gray scale of pixel
Value carries out convolution, obtains gradient information, convolution method are as follows:
Wherein, I (i, j) is the gray value of pixel (i, j), and P (i, j) and Q (i, j) respectively represent the pixel in the direction x and y
Gradient information；
S212, the gradient information according to the pixel (i, j) obtained in step S211 in the direction x and y, calculate the ladder of the pixel
Degree amplitude G (i, j) and gradient direction θ (i, j):
G (i, j) is greater than the pixel collection of preset threshold as all probable edge information of ship picture frame.
7. being based on Canny operator and the morphologic ship detecting method of Gauss as claimed in claim 5, which is characterized in that institute
State step S23 the following steps are included:
S231, strong edge threshold value T1 and weak edge threshold T2 is set, wherein T1 > T2；
S232, G (i, j) indicate the gradient amplitude of pixel (i, j), when G (i, j) is greater than T1, then by pixel (i, j) as strong
Marginal point, all strong edge pixels are all ship profile points；If T1 > G (i, j) > T2, by pixel (i, the j) conduct
Weak edge pixel point；
S233, strong edge point and weak marginal point are then connected according to eight connectivity metric, obtain the final ship profile letter of Canny operator
Breath.
8. as claimed in claim 1 or 3 be based on Canny operator and the morphologic ship detecting method of Gauss, which is characterized in that
In the step S3, comprising steps of
Ship profile in S31, the ship profile information for enabling step S2 extract is S (x, y), and β is adaptive Gaussian mixture model device
Weighting parameter, E (β) indicate the desired value of the adaptive Gaussian mixture model device, and G (x, y) indicates the adaptive Gaussian mixture model device
Operator, operator * indicate the convolution operation of G (x, y) and S (x, y), and ▽ operator is derivative operation symbol, and parameter lambda is that characterization E (β) is received
The coefficient of speed is held back, then the expression formula of E (β) are as follows:
E (β)=∫ ∫ [(S (x, y)-G (x, y) * S (x, y))2+λ((▽β)-1)2],
Acquire the minimum value of E (β)；
S32, according to the E (β) minimum value, find out the value of β, the best initial weights parameter as the adaptive Gaussian mixture model device；
Enable Sp(x, y) is the filtered ship profile of adaptive Gaussian mixture model device, and the sef-adapting filter utilizes public affairs according to the value of β
FormulaFiltering and noise reduction is further smoothed to ship profile S (x, y).
9. being based on Canny operator and the morphologic ship detecting method of Gauss as described in claim 1, which is characterized in that institute
State step S4 the following steps are included:
S41, etching operation is carried out to the ship profile information for further smoothing denoising obtained in step S3, completely removed
The edge of background object；
S42, expansive working is carried out to the ship profile information after etching operation, reconstructs true ship profile, obtains final
Ship detecting result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810967927.0A CN109215018A (en) | 2018-08-23 | 2018-08-23 | Based on Canny operator and the morphologic ship detecting method of Gauss |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201810967927.0A CN109215018A (en) | 2018-08-23 | 2018-08-23 | Based on Canny operator and the morphologic ship detecting method of Gauss |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN109215018A true CN109215018A (en) | 2019-01-15 |
Family
ID=64989176
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201810967927.0A Pending CN109215018A (en) | 2018-08-23 | 2018-08-23 | Based on Canny operator and the morphologic ship detecting method of Gauss |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109215018A (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109700435A (en) * | 2019-01-28 | 2019-05-03 | 上海得舟信息科技有限公司 | A kind of projection intravenous angiography device and its image processing method |
| CN109934269A (en) * | 2019-02-25 | 2019-06-25 | 中国电子科技集团公司第三十六研究所 | A kind of opener recognition methods of electromagnetic signal and device |
| CN110223266A (en) * | 2019-03-08 | 2019-09-10 | 湖南工业大学 | A kind of Railway wheelset tread damage method for diagnosing faults based on depth convolutional neural networks |
| CN110852317A (en) * | 2019-09-16 | 2020-02-28 | 浙江工业大学 | Small-scale target detection method based on weak edge |
| CN110852323A (en) * | 2019-11-18 | 2020-02-28 | 南京莱斯电子设备有限公司 | Angular point-based aerial target detection method |
| CN111798392A (en) * | 2020-06-29 | 2020-10-20 | 佛山科学技术学院 | Object edge noise reduction method and system for infrared image |
| CN112528868A (en) * | 2020-12-14 | 2021-03-19 | 江苏师范大学 | Illegal line pressing judgment method based on improved Canny edge detection algorithm |
| CN113255537A (en) * | 2021-06-01 | 2021-08-13 | 贵州财经大学 | Image enhancement denoising method for identifying sailing ship |
| CN115272142A (en) * | 2022-09-30 | 2022-11-01 | 南通市通州区华凯机械有限公司 | Scene image preprocessing method of immersive driving simulator |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080267470A1 (en) * | 2007-04-27 | 2008-10-30 | Three Palm Software | Algorithms for selecting mass density candidates from digital mammograms |
| CN104732536A (en) * | 2015-03-18 | 2015-06-24 | 广东顺德西安交通大学研究院 | Sub-pixel edge detection method based on improved morphology |
-
2018
- 2018-08-23 CN CN201810967927.0A patent/CN109215018A/en active Pending
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20080267470A1 (en) * | 2007-04-27 | 2008-10-30 | Three Palm Software | Algorithms for selecting mass density candidates from digital mammograms |
| CN104732536A (en) * | 2015-03-18 | 2015-06-24 | 广东顺德西安交通大学研究院 | Sub-pixel edge detection method based on improved morphology |
Non-Patent Citations (2)
| Title |
|---|
| G DENG,L W CAHILL: "An adaptive Gaussian filter for noise reduction and edge detection", 《NUCLEAR SCIENCE SYMPOSIUM & MEDICAL IMAGING CONFERENCE IEEE》 * |
| 许开宇: "基于红外图像的运动船舶检测及跟踪技术的研究", 《中国优秀博硕士学位论文全文数据库(博士)》 * |
Cited By (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109700435A (en) * | 2019-01-28 | 2019-05-03 | 上海得舟信息科技有限公司 | A kind of projection intravenous angiography device and its image processing method |
| CN109934269A (en) * | 2019-02-25 | 2019-06-25 | 中国电子科技集团公司第三十六研究所 | A kind of opener recognition methods of electromagnetic signal and device |
| CN110223266A (en) * | 2019-03-08 | 2019-09-10 | 湖南工业大学 | A kind of Railway wheelset tread damage method for diagnosing faults based on depth convolutional neural networks |
| CN110852317A (en) * | 2019-09-16 | 2020-02-28 | 浙江工业大学 | Small-scale target detection method based on weak edge |
| CN110852323A (en) * | 2019-11-18 | 2020-02-28 | 南京莱斯电子设备有限公司 | Angular point-based aerial target detection method |
| CN111798392A (en) * | 2020-06-29 | 2020-10-20 | 佛山科学技术学院 | Object edge noise reduction method and system for infrared image |
| CN112528868A (en) * | 2020-12-14 | 2021-03-19 | 江苏师范大学 | Illegal line pressing judgment method based on improved Canny edge detection algorithm |
| CN112528868B (en) * | 2020-12-14 | 2022-04-15 | 江苏师范大学 | Illegal line pressing judgment method based on improved Canny edge detection algorithm |
| CN113255537A (en) * | 2021-06-01 | 2021-08-13 | 贵州财经大学 | Image enhancement denoising method for identifying sailing ship |
| CN115272142A (en) * | 2022-09-30 | 2022-11-01 | 南通市通州区华凯机械有限公司 | Scene image preprocessing method of immersive driving simulator |
| CN115272142B (en) * | 2022-09-30 | 2023-04-18 | 南通市通州区华凯机械有限公司 | Scene image preprocessing method of immersive driving simulator |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109215018A (en) | Based on Canny operator and the morphologic ship detecting method of Gauss | |
| Gershikov et al. | Horizon line detection in marine images: which method to choose? | |
| CN103927751A (en) | Water surface optical visual image target area detection method based on gradient information fusion | |
| CN109427055B (en) | Remote sensing image sea surface ship detection method based on visual attention mechanism and information entropy | |
| US9031285B2 (en) | Detection of floating objects in maritime video using a mobile camera | |
| CN108229433B (en) | Method for detecting ship landing on shore based on straight-line segment detection and shape characteristics | |
| CN103761731A (en) | Small infrared aerial target detection method based on non-downsampling contourlet transformation | |
| CN103679694B (en) | A kind of ship small targets detection method based on panoramic vision | |
| Lipschutz et al. | New methods for horizon line detection in infrared and visible sea images | |
| CN106886747B (en) | It is a kind of based on extension wavelet transformation complex background under Ship Detection | |
| CN113902633B (en) | Sea surface search and rescue target strengthening method based on polarization information | |
| CN110728668B (en) | Airspace high-pass filter for maintaining small target form | |
| KR20210099371A (en) | The method and System of Wave Observation Using Camera Module for Ocean Observation Buoy | |
| Zhan et al. | Effective waterline detection for unmanned surface vehicles in inland water | |
| CN114821358A (en) | Optical remote sensing image marine ship target extraction and identification method | |
| Rodin et al. | Detectability of objects at the sea surface in visible light and thermal camera images | |
| Marques et al. | An algorithm for the detection of vessels in aerial images | |
| Shi et al. | Obstacle type recognition in visual images via dilated convolutional neural network for unmanned surface vehicles | |
| CN111428624B (en) | Optical remote sensing image on-orbit ship target detection method | |
| Hashmani et al. | A survey on edge detection based recent marine horizon line detection methods and their applications | |
| CN111089586B (en) | All-day star sensor star point extraction method based on multi-frame accumulation algorithm | |
| CN112669332A (en) | Method for judging sea and sky conditions and detecting infrared target based on bidirectional local maximum and peak local singularity | |
| Kocak et al. | Detection and tracking of ships using a stereo vision system | |
| CN114863373A (en) | Offshore unmanned platform monitoring method and offshore unmanned platform | |
| CN114429593A (en) | Infrared small target detection method based on rapid guided filtering and application thereof |
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 | ||
| RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190115 |