CN103902972A - Water surface moving platform visual system image analyzing and processing method - Google Patents
Water surface moving platform visual system image analyzing and processing method Download PDFInfo
- Publication number
- CN103902972A CN103902972A CN201410105371.6A CN201410105371A CN103902972A CN 103902972 A CN103902972 A CN 103902972A CN 201410105371 A CN201410105371 A CN 201410105371A CN 103902972 A CN103902972 A CN 103902972A
- Authority
- CN
- China
- Prior art keywords
- image
- target
- identified
- processing
- frame
- 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
Images
Landscapes
- Image Analysis (AREA)
- Image Processing (AREA)
Abstract
The invention belongs to the field of information processing in the information engineering field, and particularly relates to a water surface moving platform visual system image analyzing and processing method. The method includes the steps of image fog removal, graying, image stabilizing, target segmenting, binaryzation, shape characteristic extracting, texture characteristic extracting, establishing of target characteristic bases of different types, and neural network training and testing. The water surface moving platform visual system image analyzing and processing method has the advantages of being high in processing instantaneity, high in image comprehensive analyzing and processing capacity, wide in application range and the like; video image fog removal and electronic image stabilizing can be conducted on a water surface moving platform in real time under the severe weather conditions such as sea fog, and various types of targets on the water surface can be recognized.
Description
Technical field
Patent of the present invention belongs to the field of information processing in information engineering field, is specifically related to a kind of water surface movable platform vision system image analysis processing method.
Background technology
Along with the appearance of high precision, high-resolution image sensors and high-speed data acquisition and embedded processing systems, real-time video processing ability is greatly improved.Also make vision system at water surface movable platform, as also more and more extensive in the application in survey vessel, cargo ship, unmanned boat, the warship that cruises etc., become the important means of the target detection of water surface movable platform, target's feature-extraction and identification and target following.
The deficiency that the vision system using in water surface movable platform at present exists is mainly reflected in:
The autonomous analyzing and processing scarce capacity of image of vision system, is embodied in:
Due to the operation at wave and water surface movable platform, can there is shake and thicken in the video image of camera collection, can cause target image None-identified or identification error, and because this exists the multiple situations such as translation, rotation, randomized jitter in fuzzy, so the complexity of processing, current existing water surface movable platform vision system lacks this processing power;
Due to marine environment complexity, the captured video image of water surface movable platform vision system often can be subject to the impact of sea fog weather condition, causes that captured video image fuzzyly degrades, contrast is low, affects image subsequent analysis and processing.Current existing video image enhancement is processed most of for land image processing, processes research very limited for the inhomogeneous image sea fog of the water surface, and water surface movable platform vision system lacks this processing power especially;
The feature extraction of the waterborne target of current water surface movable platform and identification major part rest on the identification of water surface ship, also do not have water surface multi-class targets (as island, rock, ship etc.) recognition function.
Due to existing software algorithm and the hardware system scarce capacity merging that matches, the vision system of most of water surface movable platform is all to match with conventional computer, play the booster action of environment or targeted surveillance, afterwards judge decision-making by artificial and comprehensive analysis of other measuring equipments, be unfavorable for the real-time implementation of the autonomous processing capacity of video image.
Summary of the invention
The object of the present invention is to provide a kind of water surface movable platform vision system image analysis processing method that is suitable for water surface movable platform.
The object of the present invention is achieved like this:
1) image mist elimination: carry out mist elimination for the low target image to be identified of contrast under sea fog and strengthen sharpening processing, obtain the video image of mist elimination sharpening:
1.1) target video image to be identified under sea fog is carried out to frame difference method background extracting and obtain initial background;
1.2) the first frame of the image/video frame to be identified under current same background is carried out to the single image mist elimination processing based on atmospheric scattering physical model, deduct picture after treatment with picture before treatment and obtain the fog shade under this background;
1.3) all frame of video under current same background all deduct this fog shade and obtain the sharpening video sequence after target video enhancing recovery to be identified under current background;
1.4) context update, and return to step 1.2) continue to carry out the sharpening target video image to be identified after output mist elimination;
2) gray processing: the target video image to be identified of the sharpening that previous step image mist elimination is obtained carries out gray processing processing;
3) image surely looks like: the target image to be identified of gray processing obtained in the previous step is carried out to image electronic surely as processing, eliminate because wave and water surface movable platform motion jitter cause the impact that target image to be identified is fuzzy, obtain the target image to be identified of final sharpening:
3.1) utilize SIFT Angular Point Extracting Method to extract respectively the feature point set A of target video image reference frame image to be identified and the feature point set B of current frame image, here reference frame is the former frame of present frame, and present frame represents the current frame of video that steady picture is processed of carrying out;
3.2) utilize sequence similarity detection algorithm to 3.1) in feature point set A and feature point set B carry out Corner Feature coupling;
3.3) utilize affine model to solve the kinematic parameter of match point in feature point set A and B to the unique point after coupling;
3.4) utilize Kalman motion filtering method to obtain kinematic parameter corresponding to current picture frame normal scan to be identified;
3.5) by 3.3) the image motion parameter and 3.4 that obtains) kinematic parameter corresponding to normal scan that obtain subtract each other kinematic parameter corresponding to randomized jitter that obtains video camera, needs the kinematic parameter that present frame is compensated;
3.6) according to 3.5) the final kinematic parameter obtaining compensates accordingly to current frame image, re-executes 3.1) continue the steady picture processing of next frame target image to be identified;
4) Target Segmentation: adopt the level and smooth partitioning algorithm based on adaptive M ean-Shift to carry out cutting apart of target background to the target image to be identified of the final sharpening of previous step, extract target;
5) binaryzation: the target that previous step is extracted is carried out simple binaryzation operation, is white by target label, and background is black, obtains white target area;
6) Shape Feature Extraction: the target area that previous step binaryzation is obtained, extract its 14 shape facilities, comprise area features, flexibility feature, tight ness rating feature, convex closure feature, projection tolerance and 6 Hu invariant moment features and 3 affine invariant moment features;
7) texture feature extraction: obtain target area according to binaryzation, obtain surely looking like target area in former target image to be identified after treatment, this target area is carried out to the extraction of textural characteristics, obtain corresponding 6 textural characteristics of target;
8) set up dissimilar target feature library: target image sample process image is cut apart, binaryzation extracts its shape facility afterwards and textural characteristics obtains;
9) neural metwork training and test: the target signature being obtained by target image Sample Storehouse is input to BP neural network and trains, the neural network that obtains training, then target signature to be identified is input in training network and is identified, obtain final recognition result.
Beneficial effect of the present invention is: water surface movable platform vision system image analysis processing method has the features such as real-time, image synthesis analyzing and processing ability is strong, applied widely of processing, can under the inclement weather environment such as sea fog, carry out real-time video image mist elimination and electronic steady image processing to water surface movable platform, and can identify water surface multi-class targets.Solve due to flating that the fast moving of mobile platform causes, due to cause under water surface sea fog weather image blurring and water surface movable platform vision system image synthesis Treatment Analysis insufficiency of function that waterborne target identification form first-class causes, adaptability is not strong, and software algorithm and hardware platform merge the not strong problem of the not enough autonomous processing capacity real-time of water surface movable platform video image causing.
Accompanying drawing explanation
Fig. 1 is water surface movable platform vision system image analysis processing device general structure block diagram;
The real-time processes and displays of Fig. 2 software interface;
The algorithm block diagram of Fig. 3 water surface movable platform vision system image analysis processing method;
Fig. 4 video image enhancement processing flow chart;
Fig. 5 single image mist elimination block diagram;
Fig. 6 video strengthens treatment effect;
The Electronic Image Stabilization process flow diagram of Fig. 7 based on feature point extraction and Kalman filtering;
Fig. 8 consecutive frame compensation schematic diagram;
Fig. 9 video image surely looks like treatment effect;
Tetra-kinds of experimental data sources of Figure 10;
Figure 11 software image treatment effect.
Embodiment
Below in conjunction with accompanying drawing, the present invention is described further.
The object of this invention is to provide a kind of water surface movable platform vision system image analysis processing method that is suitable for water surface movable platform, the technical matters that emphasis solves: solve the flating that causes due to the fast moving of mobile platform, due to cause under water surface sea fog weather image blurring and water surface movable platform vision system image synthesis Treatment Analysis insufficiency of function that waterborne target identification form first-class causes, adaptability is not strong, and software algorithm and hardware platform merge the not strong problem of the not enough autonomous processing capacity real-time of water surface movable platform video image causing.Develop the real-time processing method of the vision system video image that is suitable for water surface movable platform: comprising: foreground software is processed real-time display interface and background process algorithm.Connect with real time high-speed video image acquisition based on FPGA and the real-time video processing platform of miniPC main frame and VxWorks operating system, realize hardware and software platform and merge.
Native system principle of work is to utilize the FPGA image capture module and the camera module that are arranged on water surface movable platform to carry out the collection of mobile platform ambient water Area Objects video image, by FPGA image capture module by the transmission of video images collecting to the processing of carrying out video image on the miniPC host platform of VxWorks embedded system is housed, then result after treatment is transferred in VGA display device and is shown, and control the position of servo-drive system adjusting camera, realize the omnidirectional video image of scene around water surface movable platform is obtained.Under embedded operation platform, mainly complete to the video image of FPGA image capture module transmission process, the control of servo-drive system and result is shown in display device, the video image of FPGA image capture module transmission is mainly carried out to following processing here: the electronic steady image of video image mist elimination, video image and target identification; The video image after main Graphics Processing and the result of target identification of display device.
Each several part annotation and effect in upper figure:
Waterborne target: the target around water surface movable platform in scene is the object that vision system will be identified, and comprising: ship, island, reef etc., mainly obtained by the camera on mobile platform.
Camera module: be used for obtaining water surface movable platform scene information around, what mainly obtain here is mobile platform video image around;
FPGA image capture module: control camera module and complete the water surface movable platform high speed acquisition of scene video image around, and the transmission of video images collecting is processed to embedded system platform;
Embedded system: core control and video image processor, the method for video image processing in this system is also the core technology place of this patent.Major function is to the processing of scene video image around of water surface movable platform, mainly complete the water surface movable platform scene video around that video image mist elimination under sea fog and video image electronic steady image obtain sharpening, thereby and carry out Target Segmentation, target's feature-extraction realization to the target identification in scene around water surface movable platform by the video image to after sharpening, and by thereby the control of servo-drive system is regulated to camera direction, thereby better obtain the water surface movable platform target image of scene around;
Servo-drive system: being used for driving the adjustment of vision system platform stance, is mainly to regulate the position of camera convenient to the water surface movable platform Real-time Obtaining of scene information around;
Display device: scene image and target recognition result around the water surface movable platform of sharpening after embedded system platform mist elimination, steady picture are processed are shown in real time.
FPGA image capture module on water surface movable platform and camera module by the mobile platform collecting around scene transmission of video images in embedded system platform, under embedded system, carry out video image processing according to algorithm steps as shown in Figure 3, thereby realize, real-time video image analysis is processed and target identification.
This water surface movable platform vision system image analysis processing method is mainly made up of target pre-service and detection module, characteristic extracting module and neural metwork training and the large core of test module three.Wherein the target image to be identified in algorithm block diagram is the water surface movable platform being gathered by FPGA image capture module and the camera module video image of scene around that embedded system platform receives; Target image sample is to collect the image pattern storehouse that a large amount of marine island, reef and various dissimilar ships forms.Target pre-service and detection module major function are water surface movable platform target image around to be carried out to mist elimination, steady picture, Target Segmentation be convenient to extract clarification of objective to be identified in this target image, and the great amount of samples in target image sample are carried out Target Segmentation and is convenient to extract the feature of these known target samples below.Characteristic extracting module mainly completes carries out feature extraction to the known target in target to be identified and target image sample, is convenient to these features of later use and carries out neural network recognization.Neural metwork training is to utilize BP neural network that the sample characteristics of known target in the target signature to be identified of extracting and target image Sample Storehouse is mated to Classification and Identification above with test module major function, thus final realization for the identification of scene objects around of water surface movable platform.
In algorithm flow shown in Fig. 3, in order to improve speed and the real-time of water surface movable platform vision system video image algorithm identified, the feature of the known target image pattern in target image Sample Storehouse generally just can extract before this platform comes into operation, then in the use procedure of water surface movable platform, only need to extract the target signature in the target image to be identified in scene around water surface movable platform, then target signature to be identified and known Sample Storehouse feature are input to neural network and carry out Classification and Identification.
Algorithm steps for target image processing to be identified is:
1) image mist elimination: carry out mist elimination for the low target image to be identified of contrast under sea fog and strengthen sharpening processing, obtain the video image of mist elimination sharpening:
1.1) target video image to be identified under sea fog is carried out to frame difference method background extracting and obtain initial background;
1.2) the first frame of the image/video frame to be identified under current same background is carried out to the single image mist elimination processing based on atmospheric scattering physical model, deduct picture after treatment with picture before treatment and obtain the fog shade under this background;
1.3) all frame of video under current same background all deduct this fog shade and obtain the sharpening video sequence after target video enhancing recovery to be identified under current background;
1.4) context update, and return to step 1.2) continue to carry out the sharpening target video image to be identified after output mist elimination;
2) gray processing: the target video image to be identified of the sharpening that previous step image mist elimination is obtained carries out gray processing processing;
3) image surely looks like: the target image to be identified of gray processing obtained in the previous step is carried out to image electronic surely as processing, eliminate because wave and water surface movable platform motion jitter cause the impact that target image to be identified is fuzzy, obtain the target image to be identified of final sharpening:
3.1) utilize SIFT Angular Point Extracting Method to extract respectively the feature point set A of target video image reference frame image to be identified and the feature point set B of current frame image, here reference frame is the former frame of present frame, and present frame represents the current frame of video that steady picture is processed of carrying out;
3.2) utilize sequence similarity detection algorithm to 3.1) in feature point set A and feature point set B carry out Corner Feature coupling;
3.3) utilize affine model to solve the kinematic parameter of match point in feature point set A and B to the unique point after coupling;
3.4) utilize Kalman motion filtering method to obtain kinematic parameter corresponding to current picture frame normal scan to be identified;
3.5) by 3.3) the image motion parameter and 3.4 that obtains) kinematic parameter corresponding to normal scan that obtain subtract each other kinematic parameter corresponding to randomized jitter that obtains video camera, needs the kinematic parameter that present frame is compensated;
3.6) according to 3.5) the final kinematic parameter obtaining compensates accordingly to current frame image, re-executes 3.1) continue the steady picture processing of next frame target image to be identified;
4) Target Segmentation: adopt the level and smooth partitioning algorithm based on adaptive M ean-Shift to carry out cutting apart of target background to the target image to be identified of the final sharpening of previous step, extract target;
5) binaryzation: the target that previous step is extracted is carried out simple binaryzation operation, is white by target label, and background is black, obtains white target area;
6) Shape Feature Extraction: the target area that previous step binaryzation is obtained, extract its 14 shape facilities, comprise area features, flexibility feature, tight ness rating feature, convex closure feature, projection tolerance and 6 Hu invariant moment features and 3 affine invariant moment features;
7) texture feature extraction: obtain target area according to binaryzation, obtain surely looking like target area in former target image to be identified after treatment, this target area is carried out to the extraction of textural characteristics, obtain corresponding 6 textural characteristics of target;
8) set up dissimilar target feature library: target image sample process image is cut apart, binaryzation extracts its shape facility afterwards and textural characteristics obtains;
9) neural metwork training and test: the target signature being obtained by target image Sample Storehouse is input to BP neural network and trains, the neural network that obtains training, then target signature to be identified is input in training network and is identified, obtain final recognition result.
Wherein each functions of modules is as follows:
Image mist elimination: under sea fog, video image seriously degrades, according to becoming objective fuzzy, be difficult to carry out target's feature-extraction, therefore under sea fog, utilize the defogging method capable of atmospheric scattering physical model to carry out the processing of mist elimination sharpening here, be convenient to follow-up steady picture and image dividing processing;
Image surely looks like: the water surface movable platform image that camera collection is come in the time of high-speed mobile can be fuzzy due to shake, therefore mobile platform capture video image is carried out to steady picture here and processes, and is convenient to succeeding target background separation;
Image is cut apart: adopt adaptive Mean-Shift image segmentation algorithm to carry out the separation of target background, extract target, be convenient to follow-up target's feature-extraction;
Feature extraction: the target splitting is above carried out to textural characteristics and Shape Feature Extraction, and wherein shape facility mainly comprises hu moment characteristics, affine constant torch feature and contour feature (area features, flexibility feature, tight ness rating feature, convex closure feature, projection tolerance);
Neural metwork training and test: the target signature input neural network extracting is above identified, provided recognition result.
Embodiment
1. video image mist elimination Processing Example
Water surface movable platform video image mist elimination process flow diagram is illustrated in fig. 4 shown below.Ultimate principle: to there being the sequence of video images to be identified of mist, first choose the first frame as with reference to frame, and this two field picture is carried out to the processing of mist elimination sharpening, obtain frame of video and fog shade after the mist elimination of present frame, the fog shade of trying to achieve before then the frame of video under same background being deducted just can be realized the processing of mist elimination sharpening, and can greatly reduce the mist elimination time, if frame difference method background extracting finds that change of background is larger, need again to extract video background and recalculate fog shade one time.The principle of wherein carrying out the processing of mist elimination sharpening for single width frame of video is: first treat recognition target image and cut apart, by analyzing sky provincial characteristics, obtain sky brightness estimated value, then ask for transmissivity according to dark primary priori theory, on this basis by solving atmospheric scattering physical model, thereby realize the clear processing of single image.
Specific procedure performing step is as follows:
1) to camera collection to sequence of frames of video utilize frame difference method to carry out background extracting;
2) the first frame of a string frame of video under current same background is carried out to the processing of single image mist elimination, and deduct picture after treatment with picture before treatment and obtain the fog shade under this background;
3) all frame of video under current same background all deduct this fog shade and are enhanced and restore later video sequence;
4) context update, and return to step 2 and continue to carry out, the sharpening video after mist elimination finally exported.
Single image mist elimination algorithm is described below:
In step 2, single image mist elimination algorithm block diagram used is as shown in Figure 5 above:
1) estimation of sky brightness A: adopt the medium filtering of 3*3 to carry out pre-service to picture, then carry out canny rim detection, then utilize Hough conversion to carry out straight-line detection, and extracting Hough, to convert the nose section obtaining be sea horizon, extracts bright spot and obtain sky brightness A above original image sea horizon region;
2) transmissivity is estimated: ask for transmissivity according to dark primary priori theory, it is as follows that transmissivity is asked for formula:
3) atmospheric scattering model solution: according to the target image to be identified that solves atmospheric scattering model formation I (x)=J (x) t (x)+A (1-t (x)), obtain mist elimination after please liking
wherein I (x) represents former target image to be identified, and J (x) represents the target image to be identified of the sharpening after mist elimination.
The design sketch that adopts the algorithm of design to strengthen processing to the vedio data under sea fog (panorama net (http://www.quanjing.com/)) is as follows.
2. image surely looks like Processing Example
Adopt the Electronic Image Stabilization processing based on the extraction of SIFT angle point and Kalman filtering.
Ultimate principle: utilize the method for extracting based on SIFT angle point to extract respectively Corner Feature set A, the B of present frame (current image to be identified) and reference frame (current image former frame image to be identified), then utilize SSDA algorithm to carry out the corners Matching of Corner Feature set A and B, select affine model to solve the kinematic parameter of image, then utilize the motion compensation process of Kalman filtering to compensate current frame image, thereby realize the electronic steady image processing of target image sequence to be identified.
Specific implementation step is as follows:
1) utilize SIFT Angular Point Extracting Method to extract respectively the feature point set A of reference frame image and the feature point set B of current frame image;
2) the Corner Feature point set A and the B that utilize SSDA (sequence similarity detection algorithm) to extract previous step carry out corners Matching;
3) utilize affine model to solve the kinematic parameter that mates angle point in feature point set A and B to the unique point after coupling;
4) utilize Kalman motion filtering method to obtain kinematic parameter corresponding to normal scan in each two field picture;
5) kinematic parameter that the kinematic parameter of the image (3) step being obtained is corresponding with the normal scan that step (4) obtains subtracts each other kinematic parameter corresponding to randomized jitter that obtains video camera, the kinematic parameter that present frame need to compensate;
6) according to the final kinematic parameter obtaining of (5) step, current two field picture to be identified is compensated accordingly, thereby realize the electronic steady image processing of present frame, then return to the first step and carry out the steady picture processing of next frame image to be identified, thereby the fuzzy steady picture video sequence of shake is eliminated in final output.
The method of utilizing present frame and reference frame to compensate recited above is illustrated in fig. 8 shown below:
The design sketch that adopts the Electronic Image Stabilization of design to carry out corresponding processing to the continuous multiple frames image that contains shake of water surface remote control speedboat collection is as follows.
3. the feature extraction of image multiple goal and identification embodiment
Four sources of experimental data: the first is to come from real water surface movable platform---unmanned boat real scene shooting data; The second comes from the data of network, comprises various ships, reef, island picture; The third comes from the actual measurement data of homemade water surface remote control ship; The 4th kind comes from the special image making software 3DMAX of computing machine, comprises the 3D model data of various different ships.In target sample data, typical target sample is as Figure 10.
By extracting the multiobject peripheral profile such as island, reef and ship and hu, bending moment and affine invariant moment features do not carry out multiobject feature extraction.Meanwhile, differ larger target for island, these superficial makingss of reef and ship, the textural characteristics of employing based on gray scale symbiosis square is as another remarkable factor of identification.By these feature composition target signature Sample Storehouses, adopt the BP neural network based on principal component characteristic optimization to identify water surface multiple goal, thereby obtain the multiple goal feature extraction of the simple and effective unmanned boat vision system water surface and recognition methods.
Concrete scheme performing step is as follows:
1) by target image sample with through mist elimination above with surely utilize the level and smooth dividing method of adaptive Mean-Shift to carry out image as target image to be identified after treatment to cut apart, reach the object that target context separates;
2) picture after cutting apart being carried out to binaryzation operation, is white by target label, and context marker is black;
3) (2) step target image to be identified after treatment and target sample image are carried out to the extraction of shape facility (extraction and 3 affine invariant moment features of area features, flexibility feature, tight ness rating feature, convex closure feature, projection tolerance and 6 Hu invariant moment features) according to the definition of each shape facility, and be kept in excel, so just obtain the shape facility of target image to be identified and the shape facility of target sample (sample in target image Sample Storehouse), be convenient to follow-up these data that read and carry out neural network recognization;
4) the former target sample image to the first step and target image to be identified carry out the extraction of the waterborne target textural characteristics (texture energy, texture entropy, texture gradient, texture correlation and unfavourable balance square) based on gray level co-occurrence matrixes, and characteristic is kept in excel form, obtain the textural characteristics of target image sample in the textural characteristics of target image to be identified and target sample storehouse;
5) using the training sample as neural network through the target sample feature database obtaining after step process above, adopt principal component analysis to be optimized this training sample, choose the neural network that characteristic quantity that wherein contribution rate is larger obtains training as the input of neural network;
6) in the neural network that in the target image to be identified preceding step being obtained, clarification of objective input to be identified (5) step trains, identify, thereby obtain recognition result.
Here choose reef, island and water surface remote control ship each 100 as target sample storehouse, then choose wherein 180 as training sample, 60 sample datas of every class target, be left 120 as test sample book, every class target contains 40 sample datas, utilize the method for shape facility and textural characteristics and BP neural network recognization to test, identification data is as follows:
Table 1 is based on different characteristic object recognition rate
4. software and hardware merges this embodiment
The software algorithm of exploitation is applied on homemade water surface remote control canoe, and the parameter of surface boat is as table 3.
Table 3 is made water surface ship technical parameter by oneself
Test in the pond of water surface remote control canoe body after Harbin Engineering University's main building.After 15 minutes, obtain experimental data, carry out the off-line analysis of site disposal image, can find out the real-time treatment effect of vedio data, see Figure 11.
First, when utilizing high-resolution image sensors to obtain waterborne target image, utilize FPGA to realize the high speed acquisition of image; FPGA and the communication of miniPC machine, complete the transmission of image to host computer; Under embedded operation platform, use waterborne target image enhancement processing algorithm to carry out the identification of deblurring, sharpening processing and the target of image; Utilize software interface to realize the real-time demonstration of result.
Provided a kind of figure image intensifying of water surface multi-class targets and recognizer that is applicable to water surface movable platform vision system, and algorithm mates with the fusion of hardware platform.
Gordian technique of the present invention has been an image analysis processing method that is applicable to water surface movable platform vision system, the gordian technique of utilizing in this disposal route has the pre-service in early stage such as video image sea fog Transformatin, electronic steady image processing, then utilizing Mean_Shift to cut apart has realized cutting apart of target, then the method for employing this target shape feature of extraction and textural characteristics is input in neural network and identifies, and has realized one and has extracted the algorithm process overall process of target identification from video image.
Claims (1)
1. a water surface movable platform vision system image analysis processing method, is characterized in that:
1) image mist elimination: carry out mist elimination for the low target image to be identified of contrast under sea fog and strengthen sharpening processing, obtain the video image of mist elimination sharpening:
1.1) target video image to be identified under sea fog is carried out to frame difference method background extracting and obtain initial background;
1.2) the first frame of the image/video frame to be identified under current same background is carried out to the single image mist elimination processing based on atmospheric scattering physical model, deduct picture after treatment with picture before treatment and obtain the fog shade under this background;
1.3) all frame of video under current same background all deduct this fog shade and obtain the sharpening video sequence after target video enhancing recovery to be identified under current background;
1.4) context update, and return to step 1.2) continue to carry out the sharpening target video image to be identified after output mist elimination;
2) gray processing: the target video image to be identified of the sharpening that previous step image mist elimination is obtained carries out gray processing processing;
3) image surely looks like: the target image to be identified of gray processing obtained in the previous step is carried out to image electronic surely as processing, eliminate because wave and water surface movable platform motion jitter cause the impact that target image to be identified is fuzzy, obtain the target image to be identified of final sharpening:
3.1) utilize SIFT Angular Point Extracting Method to extract respectively the feature point set A of target video image reference frame image to be identified and the feature point set B of current frame image, here reference frame is the former frame of present frame, and present frame represents the current frame of video that steady picture is processed of carrying out;
3.2) utilize sequence similarity detection algorithm to 3.1) in feature point set A and feature point set B carry out Corner Feature coupling;
3.3) utilize affine model to solve the kinematic parameter of match point in feature point set A and B to the unique point after coupling;
3.4) utilize Kalman motion filtering method to obtain kinematic parameter corresponding to current picture frame normal scan to be identified;
3.5) by 3.3) the image motion parameter and 3.4 that obtains) kinematic parameter corresponding to normal scan that obtain subtract each other kinematic parameter corresponding to randomized jitter that obtains video camera, needs the kinematic parameter that present frame is compensated;
3.6) according to 3.5) the final kinematic parameter obtaining compensates accordingly to current frame image, re-executes 3.1) continue the steady picture processing of next frame target image to be identified;
4) Target Segmentation: adopt the level and smooth partitioning algorithm based on adaptive M ean-Shift to carry out cutting apart of target background to the target image to be identified of the final sharpening of previous step, extract target;
5) binaryzation: the target that previous step is extracted is carried out simple binaryzation operation, is white by target label, and background is black, obtains white target area;
6) Shape Feature Extraction: the target area that previous step binaryzation is obtained, extract its 14 shape facilities, comprise area features, flexibility feature, tight ness rating feature, convex closure feature, projection tolerance and 6 Hu invariant moment features and 3 affine invariant moment features;
7) texture feature extraction: obtain target area according to binaryzation, obtain surely looking like target area in former target image to be identified after treatment, this target area is carried out to the extraction of textural characteristics, obtain corresponding 6 textural characteristics of target;
8) set up dissimilar target feature library: target image sample process image is cut apart, binaryzation extracts its shape facility afterwards and textural characteristics obtains;
9) neural metwork training and test: the target signature being obtained by target image Sample Storehouse is input to BP neural network and trains, the neural network that obtains training, then target signature to be identified is input in training network and is identified, obtain final recognition result.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410105371.6A CN103902972A (en) | 2014-03-21 | 2014-03-21 | Water surface moving platform visual system image analyzing and processing method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201410105371.6A CN103902972A (en) | 2014-03-21 | 2014-03-21 | Water surface moving platform visual system image analyzing and processing method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN103902972A true CN103902972A (en) | 2014-07-02 |
Family
ID=50994285
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201410105371.6A Pending CN103902972A (en) | 2014-03-21 | 2014-03-21 | Water surface moving platform visual system image analyzing and processing method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103902972A (en) |
Cited By (19)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104217215A (en) * | 2014-08-28 | 2014-12-17 | 哈尔滨工程大学 | Classification and identification method for foggy water surface image and clear water surface image |
| CN105787907A (en) * | 2016-03-25 | 2016-07-20 | 北京环境特性研究所 | Real-time video analysis system |
| CN105976330A (en) * | 2016-04-27 | 2016-09-28 | 大连理工大学 | Embedded foggy-weather real-time video image stabilization method |
| CN106156781A (en) * | 2016-07-12 | 2016-11-23 | 北京航空航天大学 | Sequence convolutional neural networks construction method and image processing method and device |
| CN106210448A (en) * | 2016-07-22 | 2016-12-07 | 恒业智能信息技术（深圳）有限公司 | A kind of video image dithering Processing for removing method |
| CN106878611A (en) * | 2016-12-29 | 2017-06-20 | 北京空间机电研究所 | A kind of intelligent reconnaissance and surveillance electro-optical system |
| CN107274723A (en) * | 2017-05-25 | 2017-10-20 | 江苏大学 | A kind of water surface ship identification of view-based access control model processing and the implementation method of early warning |
| CN108053382A (en) * | 2017-12-25 | 2018-05-18 | 北京航空航天大学 | A kind of visual characteristic defogging is surely as detection system |
| CN108388899A (en) * | 2018-01-29 | 2018-08-10 | 哈尔滨工程大学 | A kind of Underwater Image feature extracting method blended based on textural characteristics and shape feature |
| CN108399621A (en) * | 2018-03-29 | 2018-08-14 | 招商局重庆交通科研设计院有限公司 | A kind of quick identification method of engineering test specimen and system |
| CN108765469A (en) * | 2016-06-16 | 2018-11-06 | 哈尔滨理工大学 | Road monitoring random motion blurred picture Fast Restoration analogue system |
| CN109118473A (en) * | 2018-07-03 | 2019-01-01 | 深圳大学 | Angular-point detection method, storage medium and image processing system neural network based |
| CN109145830A (en) * | 2018-08-24 | 2019-01-04 | 浙江大学 | A kind of intelligence water gauge recognition methods |
| CN110619606A (en) * | 2019-09-19 | 2019-12-27 | 浙江大搜车软件技术有限公司 | Panoramic image determination method and device, computer equipment and storage medium |
| CN111369938A (en) * | 2020-04-26 | 2020-07-03 | 上海年润光电科技有限公司 | Modularized LED display screen and control system thereof |
| CN111627047A (en) * | 2020-05-20 | 2020-09-04 | 上海海洋大学 | Underwater fish dynamic visual sequence moving target detection method |
| CN113570607A (en) * | 2021-06-30 | 2021-10-29 | 北京百度网讯科技有限公司 | Target segmentation method and device and electronic equipment |
| EP3848854A4 (en) * | 2018-09-04 | 2022-06-15 | Seadronix Corp. | Method for acquiring object information and apparatus for performing same |
| CN114863373A (en) * | 2022-04-19 | 2022-08-05 | 华南理工大学 | Offshore unmanned platform monitoring method and offshore unmanned platform |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102750674A (en) * | 2012-04-26 | 2012-10-24 | 长春理工大学 | Video image defogging method based on self-adapting allowance |
| WO2013018101A1 (en) * | 2011-08-03 | 2013-02-07 | Indian Institute Of Technology, Kharagpur | Method and system for removal of fog, mist or haze from images and videos |
| CN103186887A (en) * | 2011-12-30 | 2013-07-03 | 株式会社日立制作所 | Image demisting device and image demisting method |
| CN103347171A (en) * | 2013-06-27 | 2013-10-09 | 河海大学常州校区 | Foggy day video processing system and method based on DSPs |
-
2014
- 2014-03-21 CN CN201410105371.6A patent/CN103902972A/en active Pending
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2013018101A1 (en) * | 2011-08-03 | 2013-02-07 | Indian Institute Of Technology, Kharagpur | Method and system for removal of fog, mist or haze from images and videos |
| CN103186887A (en) * | 2011-12-30 | 2013-07-03 | 株式会社日立制作所 | Image demisting device and image demisting method |
| CN102750674A (en) * | 2012-04-26 | 2012-10-24 | 长春理工大学 | Video image defogging method based on self-adapting allowance |
| CN103347171A (en) * | 2013-06-27 | 2013-10-09 | 河海大学常州校区 | Foggy day video processing system and method based on DSPs |
Non-Patent Citations (5)
| Title |
|---|
| ACHALA CHATHURANGA APONSO EL AT: ""Review on State of Art Image Enhancement and Restoration Methods for a Vision Based Driver Assistance System with De-weathering"", 《INTERNATIONAL CONFERENCE OF SOFT COMPUTING AND PATTERN RECOGNITION》 * |
| ARTUR LOZA EL AT: ""Fast Algorithm for Restoration of Foggy Images"", 《ELECTRONICS, CIRCUITS, AND SYSTEMS (ICECS), 2013 IEEE 20TH INTERNATIONAL CONFERENCE ON》 * |
| ZHONGLI MA EL AT: ""Video Image Clarity Algorithm Research"", 《LECTURE NOTES IN COMPUTER SCIENCE》 * |
| 李薇: ""基于人工神经网络的水面运动目标识别技术研究"", 《中国优秀硕士学位论文全文数据库 工程科技Ⅱ辑》 * |
| 肖飞: ""基于图像特征提取和特征点描述的匹配算法研究及其应用"", 《中国优秀硕士学位论文全文数据库 信息科技辑》 * |
Cited By (33)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104217215B (en) * | 2014-08-28 | 2017-11-28 | 哈尔滨工程大学 | A kind of classifying identification method of water surface foggy image and picture rich in detail |
| CN104217215A (en) * | 2014-08-28 | 2014-12-17 | 哈尔滨工程大学 | Classification and identification method for foggy water surface image and clear water surface image |
| CN105787907A (en) * | 2016-03-25 | 2016-07-20 | 北京环境特性研究所 | Real-time video analysis system |
| CN105976330A (en) * | 2016-04-27 | 2016-09-28 | 大连理工大学 | Embedded foggy-weather real-time video image stabilization method |
| CN105976330B (en) * | 2016-04-27 | 2019-04-09 | 大连理工大学 | A kind of embedded greasy weather real time video image stabilization |
| CN108765469B (en) * | 2016-06-16 | 2022-02-08 | 哈尔滨理工大学 | Road monitoring random motion blurred image fast restoration simulation system |
| CN108765469A (en) * | 2016-06-16 | 2018-11-06 | 哈尔滨理工大学 | Road monitoring random motion blurred picture Fast Restoration analogue system |
| US10504209B2 (en) | 2016-07-12 | 2019-12-10 | Beihang University | Image dehazing method |
| CN106156781B (en) * | 2016-07-12 | 2019-09-10 | 北京航空航天大学 | Sort convolutional neural networks construction method and its image processing method and device |
| CN106156781A (en) * | 2016-07-12 | 2016-11-23 | 北京航空航天大学 | Sequence convolutional neural networks construction method and image processing method and device |
| CN106210448B (en) * | 2016-07-22 | 2020-08-18 | 深圳市智慧城市建设运行管理有限公司 | Video image jitter elimination processing method |
| CN106210448A (en) * | 2016-07-22 | 2016-12-07 | 恒业智能信息技术（深圳）有限公司 | A kind of video image dithering Processing for removing method |
| CN106878611A (en) * | 2016-12-29 | 2017-06-20 | 北京空间机电研究所 | A kind of intelligent reconnaissance and surveillance electro-optical system |
| CN106878611B (en) * | 2016-12-29 | 2020-02-11 | 北京空间机电研究所 | Intelligent reconnaissance monitoring photoelectric system |
| CN107274723A (en) * | 2017-05-25 | 2017-10-20 | 江苏大学 | A kind of water surface ship identification of view-based access control model processing and the implementation method of early warning |
| CN108053382A (en) * | 2017-12-25 | 2018-05-18 | 北京航空航天大学 | A kind of visual characteristic defogging is surely as detection system |
| CN108388899B (en) * | 2018-01-29 | 2022-03-01 | 哈尔滨工程大学 | Underwater sound image feature extraction method based on fusion of texture features and shape features |
| CN108388899A (en) * | 2018-01-29 | 2018-08-10 | 哈尔滨工程大学 | A kind of Underwater Image feature extracting method blended based on textural characteristics and shape feature |
| CN108399621B (en) * | 2018-03-29 | 2023-10-13 | 招商局重庆交通科研设计院有限公司 | Engineering test piece rapid identification method and system |
| CN108399621A (en) * | 2018-03-29 | 2018-08-14 | 招商局重庆交通科研设计院有限公司 | A kind of quick identification method of engineering test specimen and system |
| CN109118473A (en) * | 2018-07-03 | 2019-01-01 | 深圳大学 | Angular-point detection method, storage medium and image processing system neural network based |
| CN109145830B (en) * | 2018-08-24 | 2020-08-25 | 浙江大学 | Intelligent water gauge identification method |
| CN109145830A (en) * | 2018-08-24 | 2019-01-04 | 浙江大学 | A kind of intelligence water gauge recognition methods |
| US11702175B2 (en) | 2018-09-04 | 2023-07-18 | Seadronix Corp. | Method for acquiring object information and apparatus for performing same |
| EP3848854A4 (en) * | 2018-09-04 | 2022-06-15 | Seadronix Corp. | Method for acquiring object information and apparatus for performing same |
| CN110619606A (en) * | 2019-09-19 | 2019-12-27 | 浙江大搜车软件技术有限公司 | Panoramic image determination method and device, computer equipment and storage medium |
| CN111369938A (en) * | 2020-04-26 | 2020-07-03 | 上海年润光电科技有限公司 | Modularized LED display screen and control system thereof |
| CN111627047B (en) * | 2020-05-20 | 2023-06-16 | 上海海洋大学 | Underwater fish dynamic visual sequence moving target detection method |
| CN111627047A (en) * | 2020-05-20 | 2020-09-04 | 上海海洋大学 | Underwater fish dynamic visual sequence moving target detection method |
| CN113570607A (en) * | 2021-06-30 | 2021-10-29 | 北京百度网讯科技有限公司 | Target segmentation method and device and electronic equipment |
| CN113570607B (en) * | 2021-06-30 | 2024-02-06 | 北京百度网讯科技有限公司 | Target segmentation method and device and electronic equipment |
| CN114863373A (en) * | 2022-04-19 | 2022-08-05 | 华南理工大学 | Offshore unmanned platform monitoring method and offshore unmanned platform |
| CN114863373B (en) * | 2022-04-19 | 2024-06-04 | 华南理工大学 | Marine unmanned platform monitoring method and marine unmanned platform |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103902972A (en) | Water surface moving platform visual system image analyzing and processing method | |
| CN110443827B (en) | Unmanned aerial vehicle video single-target long-term tracking method based on improved twin network | |
| CN111626290B (en) | Infrared ship target detection and identification method under complex sea surface environment | |
| CN105023278B (en) | A kind of motion target tracking method and system based on optical flow method | |
| CN104933680B (en) | A kind of intelligent quick sea fog minimizing technology of unmanned boat vision system video | |
| CN107808161B (en) | Underwater target identification method based on optical vision | |
| Song et al. | Integrate MSRCR and mask R-CNN to recognize underwater creatures on small sample datasets | |
| CN108053382B (en) | A kind of visual characteristic defogging is steady as detection system | |
| CN109919026B (en) | Surface unmanned ship local path planning method | |
| Lipschutz et al. | New methods for horizon line detection in infrared and visible sea images | |
| Petraglia et al. | Pipeline tracking and event classification for an automatic inspection vision system | |
| CN109978799A (en) | A kind of maritime affairs UAV Video image defogging method based on deep learning | |
| Sun et al. | IRDCLNet: Instance segmentation of ship images based on interference reduction and dynamic contour learning in foggy scenes | |
| Liu et al. | A robust vision-based skyline detection algorithm under different weather conditions | |
| Tuominen et al. | Cloud detection and movement estimation based on sky camera images using neural networks and the Lucas-Kanade method | |
| Tarchoun et al. | Hand-Crafted Features vs Deep Learning for Pedestrian Detection in Moving Camera. | |
| CN115410039A (en) | Coal foreign matter detection system and method based on improved YOLOv5 algorithm | |
| CN109558877B (en) | KCF-based offshore target tracking algorithm | |
| CN115007474A (en) | Coal dressing robot and coal dressing method based on image recognition | |
| Yang et al. | A review of intelligent ship marine object detection based on RGB camera | |
| Wang et al. | An adaptive and online underwater image processing algorithm implemented on miniature biomimetic robotic fish | |
| Xing et al. | Image-enhanced YOLOv5 and deep sort underwater multi-moving target tracking method | |
| Tseng et al. | Blind-spot vehicle detection using motion and static features | |
| CN116246139A (en) | Target identification method based on multi-sensor fusion for unmanned ship navigation environment | |
| CN116310892A (en) | Marine rescue method based on improved YOLOV4 target detection |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into 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: 20140702 |