CN110472608A - Image recognition tracking processing method and system - Google Patents
Image recognition tracking processing method and system Download PDFInfo
- Publication number
- CN110472608A CN110472608A CN201910774328.1A CN201910774328A CN110472608A CN 110472608 A CN110472608 A CN 110472608A CN 201910774328 A CN201910774328 A CN 201910774328A CN 110472608 A CN110472608 A CN 110472608A
- Authority
- CN
- China
- Prior art keywords
- template
- image
- frame
- picture
- matching
- 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
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/74—Image or video pattern matching; Proximity measures in feature spaces
- G06V10/75—Organisation of the matching processes, e.g. simultaneous or sequential comparisons of image or video features; Coarse-fine approaches, e.g. multi-scale approaches; using context analysis; Selection of dictionaries
- G06V10/751—Comparing pixel values or logical combinations thereof, or feature values having positional relevance, e.g. template matching
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/41—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items
- G06V20/42—Higher-level, semantic clustering, classification or understanding of video scenes, e.g. detection, labelling or Markovian modelling of sport events or news items of sport video content
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/40—Scenes; Scene-specific elements in video content
- G06V20/46—Extracting features or characteristics from the video content, e.g. video fingerprints, representative shots or key frames
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/16—Human faces, e.g. facial parts, sketches or expressions
- G06V40/172—Classification, e.g. identification
Abstract
The present invention provides a kind of image recognition tracking processing method and system, and the method comprising the steps of: obtaining current video, and carries out sub-frame processing to the current video;The selection of target frame is carried out to first frame image, to be set as feature templates, and target's feature-extraction is carried out to the picture of each frame of the current video;Template matching is carried out to the picture after extraction, and template library is updated according to matching result;It controls template library output matching result, and when determining the matching result and meeting pre-set level condition, exports the corresponding video frame images of the matching result.The present invention is by carrying out the target's feature-extraction, the judgement for carrying out template matching, being updated to template library and carrying out pre-set level condition design, improve the precision of target detection, have compressed the time of detection, the accuracy and recognition efficiency of identification are improved, and is further ensured that tracking to the accuracy and robustness to set the goal.
Description
Technical field
The present invention relates to electronic information and technical field of computer vision more particularly to a kind of image recognition to track processing side
Method and system.
Background technique
Image procossing is a very popular at present research field, it is well by computer vision and artificial intelligence knot
It has been closed that, be the Important Platform of AI Yu signal processing industrialization.And in image procossing, the recognition and tracking of target is always
The direction they people research and explored.
Specifically, there are mainly two types of algorithmic systems for target identification (+tracking): first is that by computer vision, computer
Graphics, the methods of Digital Image Processing handle image, analyze, and convert, and extract, to complete correlation function;Second is that
With advanced AI algorithm, such as machine learning, deep learning (neural network) constructs a physical-statistical model, constantly to it
Middle received image signal sample is trained and assesses, and is to the model to set the goal until generating one and can be identified with efficiently and accurately
Only.Certainly both methods also often fusion, cross-reference now, to promote recognition performance and efficiency.
In traditional computer vision or digital image processing method use process, algorithm and process are mostly more lonely
It is vertical, it cannot simply merge, and be unable to speed up processing using traditional algorithm, processing accuracy cannot be improved, be more pair
A kind of realization of original theory, for identifying that demanding image or its identification precision of target are low, recognition efficiency is low.
Summary of the invention
The technical problem to be solved by the present invention is to provide a kind of identification high image recognition tracking processing method of precision and it is
System.
In order to solve the above technical problems, the image recognition tracking processing method that invention provides, the method includes the steps:
Current face's data are obtained, and carry out account identification according to current face's data, to obtain target account;
Current video is obtained, and sub-frame processing is carried out to the current video;
The selection of target frame is carried out to first frame image, to be set as feature templates, and to each frame of the current video
Picture carries out target's feature-extraction;
Template matching is carried out to the picture after extraction, and template library is updated according to matching result;
Control template library output matching result, and when determining the matching result and meeting pre-set level condition,
Export the corresponding video frame images of the matching result.
Preferably, the picture to each frame of the current video carries out the step of target's feature-extraction and includes:
The picture is converted into grayscale image, and carries out Gradient Features extraction in the horizontal and vertical directions respectively;
The grayscale image is converted picture gradient by the absolute value that the gradient value on each pixel is taken in the picture
Figure, and gaussian filtering and normalized are sequentially carried out to the picture gradient map;
The Gradient Features of template image are extracted in the horizontal and vertical directions respectively, are taken in the template image
The absolute value of gradient value on each pixel;
Template gradient map is converted by the template image, and gaussian filtering process is carried out to the template gradient map;
The obtained template gradient map is carried out expanding calculation process, then is normalized and thresholding processing.
Preferably, for the first frame image, described pair extract after picture the step of carrying out template matching packet
It includes:
It controls the target frame and carries out Object selection, and the picture is split according to external frame;
Generate monitoring standard module and according to the size of target dynamic adjustment search box size.
Preferably, for the nth frame image of the current video, the picture after described pair of extraction carries out template matching
The step of include:
The nth frame image is compareed with (N-1) frame image, to obtain the average image under present frame;
It is handled according to the average image matching degree, to obtain matching value, and the matching value and matching threshold is carried out
Compare;
By treated, the average image is matched with (N-1) frame image;
When successful match, the matching result is exported, when matching failed, is continued and template library progress
Match.
Preferably, for the nth frame image of the current video, the picture after described pair of extraction carries out template matching
The step of include:
The nth frame image is matched with the template library, to obtain the detecting and tracking result figure of present frame;
The template library is updated according to the detecting and tracking result figure, and is carried out with the updated template library
Matching;
When successful match, each template time state saved by the template library works as matching to judge operating status
When failed, search range is expanded according to linearly.
Preferably, described pair extract after the picture carry out template matching the step of include:
Transformation tracking box is search box, and API is called to be matched;
Minimum position and maximum position are modified, and the tracking box is updated, while by the tracking box
It is passed to the template library as template, so that as matched correlate template next time;
It is matched with the template library, judges the motion state of current goal, and institute is changed according to the motion state
State the size of search box.
Preferably, described pair extract after the picture carry out template matching the step of include:
When determining current goal loss, judge whether capture apparatus occurs zoom;
When determining the capture apparatus generation zoom, amplifying parameters are incoming, and adjudicate and obscure for image;
When determining the capture apparatus zoom not occurring, it is determined as decoy.
Preferably, after described the step of being determined as decoy, the method also includes:
Point target is converted by Area Objects, tracks function into the point target, and prediction locus is so that positioning target position
It sets.
Compared with the relevant technologies, image recognition tracking processing method provided by the invention has the following beneficial effects: logical
It crosses and carries out the target's feature-extraction, carry out template matching, template library is updated and carries out the pre-set level condition
Judgement design, improves the precision of target detection, has compressed the time of detection, improve the accuracy and recognition efficiency of identification,
And it is further ensured that tracking to the accuracy and robustness to set the goal.
The another object of the embodiment of the present invention is to provide a kind of image recognition tracking processing system, comprising:
Sub-frame processing module carries out sub-frame processing for obtaining current video, and to the current video;
Characteristic extracting module, for carrying out the selection of target frame to first frame image, to be set as feature templates, and to described
The picture of each frame of current video carries out target's feature-extraction;
Template renewal module for carrying out template matching to the picture after extraction, and updates mould according to matching result
Plate library;
Image output module, for controlling the template library output matching result, and it is full when determining the matching result
When sufficient pre-set level condition, the corresponding video frame images of the matching result are exported.
Preferably, the characteristic extracting module is also used to:
The picture is converted into grayscale image, and carries out Gradient Features extraction in the horizontal and vertical directions respectively;
The grayscale image is converted picture gradient by the absolute value that the gradient value on each pixel is taken in the picture
Figure, and gaussian filtering and normalized are sequentially carried out to the picture gradient map;
The Gradient Features of template image are extracted in the horizontal and vertical directions respectively, are taken in the template image
The absolute value of gradient value on each pixel;
Template gradient map is converted by the template image, and gaussian filtering process is carried out to the template gradient map;
The obtained template gradient map is carried out expanding calculation process, then is normalized and thresholding processing.
Detailed description of the invention
Fig. 1 is the flow diagram for the image recognition tracking processing method that first embodiment of the invention provides;
Fig. 2 is the flow diagram for the image recognition tracking processing method that second embodiment of the invention provides;
Fig. 3 is the structural schematic diagram that the image recognition that third embodiment of the invention provides tracks processing system.
Specific embodiment
In order to make the objectives, technical solutions, and advantages of the present invention clearer, with reference to the accompanying drawings and embodiments, right
The present invention is further elaborated.It should be appreciated that the specific embodiments described herein are merely illustrative of the present invention, and
It is not used in the restriction present invention.
Embodiment one
Referring to Fig. 1, being the flow diagram for the image recognition tracking processing method that first embodiment of the invention provides, packet
Include step:
Step S10 obtains current video, and carries out sub-frame processing to the current video;
Wherein, after video flowing is inputted by photographic equipment, progress sub-frame processing first, to obtain multiple video frames;
Step S20 carries out the selection of target frame to first frame image, to be set as feature templates, and to the current video
The picture of each frame carries out target's feature-extraction;
Wherein, by carrying out the selection of target frame, to facilitate the setting of template, target signature is carried out to the picture of each frame
Extraction process is specifically divided into manually and automatically two kinds;Then to treated, picture carries out template matching
Step S30 carries out template matching to the picture after extraction, and updates template library according to matching result;
Wherein, template library is that same database connects, and new template can be all generated after having matched each time;
Step S40 controls the template library output matching result, and meets pre-set level when determining the matching result
When condition, the corresponding video frame images of the matching result are exported;
Wherein, if as a result meeting index, export this frame in video, if do not met, transfer monitoring template into
Row target detection, if it is subsequent determine meet index if can equally export this frame, if it could not, assertive goal has disappeared
It loses, program stopped;
Preferably, transmission of video finishes program and stops by force, and video is a generalized concept in the present embodiment, can both refer to
Static image is continuously introduced into, and can also refer to the video pictures of video or camera shooting.
In the present embodiment, by carrying out the target's feature-extraction, progress template matching, being updated and go forward side by side to template library
The judgement design of the row pre-set level condition, improves the precision of target detection, has compressed the time of detection, improve identification
Accuracy and recognition efficiency, and be further ensured that tracking to the accuracy and robustness that set the goal.
Embodiment two
Referring to Fig. 2, being the flow diagram for the image recognition tracking processing method that second embodiment of the invention provides, packet
Include step:
Step S11 obtains current video, and carries out sub-frame processing to the current video;
Wherein, after video flowing is inputted by photographic equipment, progress sub-frame processing first, to obtain multiple video frames;
Step S21 carries out the selection of target frame to first frame image, to be set as feature templates, and to the current video
The picture of each frame is converted to grayscale image, and carries out Gradient Features extraction in the horizontal and vertical directions respectively;
Wherein, the extraction of the Gradient Features is carried out by using the mode that small kernel extracts, and in the step, by adopting
With the Gradient Features of sobel operator extraction image;
Step S31 takes the absolute value of the gradient value on each pixel in the picture, converts the grayscale image to
Picture gradient map, and gaussian filtering and normalized are sequentially carried out to the picture gradient map;
Wherein, by carrying out the design of gaussian filtering and normalized, so as to effectively remove the picture gradient map
In noise effect;
Step S41 respectively in the horizontal and vertical directions extracts the Gradient Features of template image, in the template
The absolute value of the gradient value on each pixel is taken in image;
Step S51 converts template gradient map for the template image, and carries out gaussian filtering to the template gradient map
Processing, carries out expanding calculation process to the obtained template gradient map, then is normalized and thresholding processing;
Wherein, how much needs of threshold value manually adjust, the purpose of the step be in order to the aspect ratio of template more thoroughly and
Complete extraction comes out, and strengthens the feature of template and weakens the feature of full image moderately to preferably position target in the picture
Position;
Step S61 carries out template matching to the picture after extraction, and updates template library according to matching result;
Wherein, for the first frame image, described pair extract after picture the step of carrying out template matching include:
It controls the target frame and carries out Object selection, and the picture is split according to external frame;
Generate monitoring standard module and according to the size of target dynamic adjustment search box size.
Preferably, for the nth frame image of the current video, the picture after described pair of extraction carries out template matching
The step of include:
The nth frame image is compareed with (N-1) frame image, to obtain the average image under present frame;
It is handled according to the average image matching degree, to obtain matching value, and the matching value and matching threshold is carried out
Compare, wherein in the step, the method used is that thresholding-expansion-corrosion-image data position is matched with operation-dynamic corrections
Degree-and threshold value comparison.Matching degree is measured according to the intensity of the profile diagram of generation;
By treated, the average image is matched with (N-1) frame image;
When successful match, the matching result is exported, when matching failed, is continued and template library progress
Match.
In addition, in the present embodiment, for the nth frame image of the current video, described pair extract after the picture into
The step of row template matching includes:
The nth frame image is matched with the template library, to obtain the detecting and tracking result figure of present frame;
The template library is updated according to the detecting and tracking result figure, and is carried out with the updated template library
Matching;
When successful match, each template time state saved by the template library works as matching to judge operating status
When failed, search range is expanded according to linearly.
Further, in this embodiment the picture after described pair of extraction includes: the step of carrying out template matching
Transformation tracking box is search box, and API is called to be matched;
Minimum position and maximum position are modified, and the tracking box is updated, while by the tracking box
It is passed to the template library as template, so that as matched correlate template next time;
It is matched with the template library, judges the motion state of current goal, and institute is changed according to the motion state
State the size of search box.
Further, described pair extract after the picture carry out template matching the step of include:
When determining current goal loss, judge whether capture apparatus occurs zoom;
When determining the capture apparatus generation zoom, amplifying parameters are incoming, and adjudicate and obscure for image;
When determining the capture apparatus zoom not occurring, it is determined as decoy.
In the present embodiment, after described the step of being determined as decoy, the method also includes:
Point target is converted by Area Objects, tracks function into the point target, and prediction locus is so that positioning target position
It sets, i.e., in the present embodiment, in the present embodiment, when there is the case where target is lost suddenly in the image of a certain frame, executes this
Algorithm:
Lose judgement 1. losing: processed image is in by two kinds of matching process, tracking box and pattern plate bolster and search box
All fail to exactly match, matched principle is according to step S51;
2. confirmation enters a decision mechanism after losing, the method used in the present invention is hardware apriority: reading equipment
Parameter, if there is the zoom of picture pick-up device, amplifying parameters are incoming, directly adjudicate and obscure for image;Otherwise judgement is decoy
3. second judgement mechanism+solution: if determination is zoom problem, doing delay process, and execute following block diagram and calculate
Method;If determination is decoy, then point target, inlet point target tracking function point_detection are converted by Area Objects
(argv), prediction locus positions target position.
Step S71 controls the template library output matching result, and meets pre-set level when determining the matching result
When condition, the corresponding video frame images of the matching result are exported;
In addition, mainly including three entity actions in template database in the present embodiment:
(1) the template created: when can be first frame frame and selecting target, automatically into the template of template library, which is designed
For a variety of gray values, several template Mat variables of more sizes are also possible to manual frame choosing creation new template, name in nth frame
Title is unified for Box
(2) more new template: monitoring template supBox and tracking template trackBox are divided into, tracking template is when matching every time
The template automatically generated, there are two sources for monitoring template: first is that the copy template that the choosing of manual frame generates, second is that being transferred from template library
Template
(3) search box size is saved.
The present embodiment has the advantage that
1. feature height reducing, precision.The present embodiment has more efficiently played gradient operator and grayscale image and has mentioned
The effect in feature is taken, by multiple-stage filtering, multistage edge extracting, the methods of template and original image processing accuracy differentiation,
It ensure that the feature extracted is more accurate accurate.This is one of innovative point of the invention.
2. processing localization, detailed-oriented.The present embodiment has abandoned the detection in traditional image procossing based on statistical law
Method will be layered using the thought for being similar to SURF algorithm in image data transformation to other dimensions (gradient+gray scale)
Secondaryization processing, and the details to template, scale and edge are analyzed, and are constrained image detection zone with template characteristic, are handled in this way
Better effect can be obtained than relying on image to deacclimatize template in traditional template matching method.This is innovative point of the invention
Two.
3. algorithm centripetalization, can regression it is strong.In actual match, the present embodiment uses frame difference+thresholding combination
Method makes full use of the correlation between picture frame and frame, and target's center may be deviateed by not only improving feature extraction step
Problem, and to detect and drawing objective contour, search range can be made constantly to return to target actual position.And this calculation
Method eliminates the influence of many inessential noises, and the noiseproof feature than being directly trained using convolutional neural networks is stronger, from
The accurate location of target so can be also collected more accurately.This is the three of innovative point of the present invention.
4. process flow simplification, programmed readability is good.The present embodiment can be rapidly to every on actual test platform
One frame picture is dealt with, and target position is collected.And program carries out secondary development directly on opencv, simplifies data knot
Structure, enhancing are readable
5. significantly improving the loss and entanglement problem of target, the anti-interference ability of algorithm and system is improved.The present embodiment
By the improvement in dynamic corrections template library and adjustment search range and algorithm above-mentioned, in actual test, reduce
It is fitted on decoy, barrier is encountered and can not find when can not be properly positioned former target and lens blur target these three situations and go out
Existing probability improves the quality and efficiency of image procossing, reduces loss caused by wrong identification.
6. configuring simplification.The present embodiment when designing program to read image data as principal function, feature extraction,
Match, template generation, lose judgement as submodule, and parametric method is passed using pointer or reference, so that program structure is simply certainly
So, readable high.And this method can be used in deep learning test data module and be used as preprocessing module before, improve net
The accuracy of network training.
Embodiment three
Referring to Fig. 3, being the structural representation for the image recognition tracking processing system 100 that third embodiment of the invention provides
Figure, comprising: sub-frame processing module 10, characteristic extracting module 11, template renewal module 12 and image output module 13, in which:
Sub-frame processing module 10 carries out sub-frame processing for obtaining current video, and to the current video
Characteristic extracting module 11, for carrying out the selection of target frame to first frame image, to be set as feature templates, and to institute
The picture for stating each frame of current video carries out target's feature-extraction.
Wherein, the characteristic extracting module 11 is also used to: the picture being converted to grayscale image, and respectively horizontal and perpendicular
Histogram carries out Gradient Features extraction upwards;The absolute value that the gradient value on each pixel is taken in the picture, by the ash
Degree figure is converted into picture gradient map, and sequentially carries out gaussian filtering and normalized to the picture gradient map;Respectively in water
The Gradient Features of template image are extracted on gentle vertical direction, the ladder on each pixel is taken in the template image
The absolute value of angle value;Template gradient map is converted by the template image, and the template gradient map is carried out at gaussian filtering
Reason;The obtained template gradient map is carried out expanding calculation process, then is normalized and thresholding processing.
Template renewal module 12 for carrying out template matching to the picture after extraction, and is updated according to matching result
Template library.
Wherein, the template renewal module 12 is also used to: being controlled the target frame and is carried out Object selection, and according to external frame
The picture is split;Generate monitoring standard module and according to the size of target dynamic adjustment search box size.
Preferably, the template renewal module 12 is also used to: the nth frame image and (N-1) frame image are carried out pair
According to obtain the average image under present frame;It is handled according to the average image matching degree, to obtain matching value, and will be described
Matching value is compared with matching threshold;By treated, the average image is matched with (N-1) frame image;When
When successful match, the matching result is exported, when matching failed, continuation is matched with the template library.
Further, the template renewal module 12 is also used to: by the nth frame image and template library progress
Match, to obtain the detecting and tracking result figure of present frame;The template library is updated according to the detecting and tracking result figure, and
It is matched with the updated template library;When successful match, each template time state for being saved by the template library
To judge operating status, when matching failed, search range is expanded according to linearly.
In addition, the template renewal module 12 is also used to: transformation tracking box is search box, and API is called to be matched;It is right
Minimum position and maximum position are modified, and are updated to the tracking box, while being passed the tracking box as template
Enter the template library, so that as matched correlate template next time;It is matched with the template library, judges current goal
Motion state, and according to the size of motion state change described search frame.
Further, the template renewal module 12 is also used to: when determining current goal loss, judging that shooting is set
It is standby whether zoom to occur;When determining the capture apparatus generation zoom, amplifying parameters are incoming, and adjudicate and obscure for image;
When determining the capture apparatus zoom not occurring, it is determined as decoy.
In the present embodiment, the template renewal module 12 is also used to: point target is converted by Area Objects, into described mesh
Mark tracking function, and prediction locus is so that positioning target position
Image output module 13 for controlling the template library output matching result, and is worked as and determines the matching result
When meeting pre-set level condition, the corresponding video frame images of the matching result are exported.
In the present embodiment, 1. feature height reducing, precision are had the advantage that.The present embodiment is more efficiently sent out
It has waved gradient operator and grayscale image and has extracted the effect in feature, by multiple-stage filtering, multistage edge extracting, template and original graph
As the methods of processing accuracy differentiation, it ensure that the feature extracted is more accurate accurate.This is one of innovative point of the invention.
2. processing localization, detailed-oriented.The present embodiment has abandoned the detection method in traditional image procossing based on statistical law, uses
Similar to the thought of SURF algorithm, change by different level will be carried out in image data transformation to other dimensions (gradient+gray scale) and handled,
And the details to template, scale and edge are analyzed, and constrain image detection zone with template characteristic, processing is than traditional mould in this way
Better effect can be obtained by deacclimatizing template by image in plate matching process.This is the two of innovative point of the invention.3. calculating
The normal direction heart, can regression it is strong.In actual match, the present embodiment uses frame difference+thresholding combination method and makes full use of
Correlation between picture frame and frame not only improves the problem of feature extraction step may deviate target's center, Er Qiewei
Detection and drafting objective contour, can be such that search range constantly returns to target actual position.And this algorithm eliminates very
The influence of mostly inessential noise, the noiseproof feature than being directly trained using convolutional neural networks is stronger, naturally also can be more
Add the accurate location for accurately collecting target.This is the three of innovative point of the present invention.4. process flow simplification, programmed readability
It is good.The present embodiment can rapidly deal with to each frame picture on actual test platform, collect target position.And
Program carries out secondary development directly on opencv, simplifies data structure, enhancing readability 5. significantly improve target loss and
Entanglement problem improves the anti-interference ability of algorithm and system.The present embodiment passes through dynamic corrections template library and adjustment search model
Enclose and algorithm above-mentioned on improvement reduce in actual test and be matched to decoy, encountering barrier can not be correct
The probability that can not find the appearance of these three situations of target when navigating to former target and lens blur, improves the quality of image procossing
And efficiency, reduce loss caused by wrong identification.6. configuring simplification.The present embodiment is when designing program to read picture number
According to as principal function, feature extraction, matching, template generation loses judgement as submodule, and pointer or reference is used to pass ginseng
Number method, so that program structure is unsophisticated, it is readable high.And by this method can be used in deep learning test data module it
It is preceding to be used as preprocessing module, improve the accuracy of network training.
The present embodiment also provides a kind of image recognition tracking processing apparatus, including storage equipment and processor, described to deposit
Storage equipment is for storing computer program, and the processor runs the computer program so that institute's image recognition tracking processing dress
It sets and executes above-mentioned image recognition tracking processing method.
The present embodiment additionally provides a kind of storage medium, and being stored thereon in above-mentioned image recognition tracking processing apparatus is made
Computer program, the program when being executed, include the following steps:
Current video is obtained, and sub-frame processing is carried out to the current video;
The selection of target frame is carried out to first frame image, to be set as feature templates, and to each frame of the current video
Picture carries out target's feature-extraction;
Template matching is carried out to the picture after extraction, and template library is updated according to matching result;
Control template library output matching result, and when determining the matching result and meeting pre-set level condition,
Export the corresponding video frame images of the matching result.The storage medium, such as: ROM/RAM, magnetic disk, CD.
It is apparent to those skilled in the art that for convenience and simplicity of description, only with above-mentioned each function
The division progress of unit, module can according to need and for example, in practical application by above-mentioned function distribution by different function
Energy unit or module are completed, i.e., the internal structure of storage device is divided into different functional unit or module, more than completing
The all or part of function of description.Each functional unit in embodiment, module can integrate in one processing unit,
It can be each unit to physically exist alone, can also be integrated in one unit with two or more units, it is above-mentioned integrated
Unit both can take the form of hardware realization, can also realize in the form of software functional units.In addition, each function list
Member, the specific name of module are also only for convenience of distinguishing each other, the protection scope being not intended to limit this application.
It will be understood by those skilled in the art that composed structure shown in Fig. 3 is not constituted to image recognition of the invention
The restriction for tracking processing system may include perhaps combining certain components or different than illustrating more or fewer components
Component layout, and the image recognition tracking processing method in Fig. 1-2 also uses more or fewer components shown in Fig. 3, or
Person combines certain components or different component layouts to realize.The so-called unit of the present invention, module etc. refer to that one kind can be by
The performed simultaneously function of processor (not shown) in image recognition tracking processing system enough completes the series of computation of specific function
Machine program can be stored in image recognition and track in the storage equipment (not shown) of processing system.
The foregoing is merely illustrative of the preferred embodiments of the present invention, is not intended to limit the invention, all in essence of the invention
Made any modifications, equivalent replacements, and improvements etc., should all be included in the protection scope of the present invention within mind and principle.
Claims (10)
1. a kind of image recognition tracking processing method, which is characterized in that the method includes the steps:
Current video is obtained, and sub-frame processing is carried out to the current video;
The selection of target frame is carried out to first frame image, to be set as feature templates, and to the picture of each frame of the current video
Carry out target's feature-extraction;
Template matching is carried out to the picture after extraction, and template library is updated according to matching result;
Control template library output matching result, and when determining the matching result and meeting pre-set level condition, output
The corresponding video frame images of the matching result.
2. image recognition tracking processing method according to claim 1, which is characterized in that described every to the current video
The picture of one frame carries out the step of target's feature-extraction and includes:
The picture is converted into grayscale image, and carries out Gradient Features extraction in the horizontal and vertical directions respectively;
The grayscale image is converted picture gradient map by the absolute value that the gradient value on each pixel is taken in the picture,
And gaussian filtering and normalized are sequentially carried out to the picture gradient map;
The Gradient Features of template image are extracted in the horizontal and vertical directions respectively, are taken in the template image each
The absolute value of gradient value on pixel;
Template gradient map is converted by the template image, and gaussian filtering process is carried out to the template gradient map;
The obtained template gradient map is carried out expanding calculation process, then is normalized and thresholding processing.
3. image recognition tracking processing method according to claim 1, which is characterized in that it is directed to the first frame image,
Described pair extract after the picture carry out template matching the step of include:
It controls the target frame and carries out Object selection, and the picture is split according to external frame;
Generate monitoring standard module and according to the size of target dynamic adjustment search box size.
4. image recognition tracking processing method according to claim 1, which is characterized in that for the of the current video
N frame image, described pair extract after the picture carry out template matching the step of include:
The nth frame image is compareed with (N-1) frame image, to obtain the average image under present frame;
It is handled according to the average image matching degree, to obtain matching value, and the matching value is compared with matching threshold;
By treated, the average image is matched with (N-1) frame image;
When successful match, the matching result is exported, when matching failed, continuation is matched with the template library.
5. image recognition tracking processing method according to claim 1, which is characterized in that for the of the current video
N frame image, described pair extract after the picture carry out template matching the step of include:
The nth frame image is matched with the template library, to obtain the detecting and tracking result figure of present frame;
The template library is updated according to the detecting and tracking result figure, and is carried out with the updated template library
Match;
When successful match, each template time state saved by the template library to judge operating status, when matching not at
When function, search range is expanded according to linearly.
6. image recognition tracking processing method according to claim 1, which is characterized in that the figure after described pair of extraction
Piece carry out template matching the step of include:
Transformation tracking box is search box, and API is called to be matched;
Minimum position and maximum position are modified, and the tracking box is updated, at the same using the tracking box as
Template is passed to the template library, so that as matched correlate template next time;
It is matched with the template library, judges the motion state of current goal, and search according to motion state change
The size of rope frame.
7. image recognition tracking processing method according to claim 1, which is characterized in that the figure after described pair of extraction
Piece carry out template matching the step of include:
When determining current goal loss, judge whether capture apparatus occurs zoom;
When determining the capture apparatus generation zoom, amplifying parameters are incoming, and adjudicate and obscure for image;
When determining the capture apparatus zoom not occurring, it is determined as decoy.
8. image recognition tracking processing method according to claim 7, which is characterized in that the step for being determined as decoy
After rapid, the method also includes:
Point target is converted by Area Objects, tracks function into the point target, and prediction locus is so that positioning target position.
9. a kind of image recognition tracks processing system characterized by comprising
Sub-frame processing module carries out sub-frame processing for obtaining current video, and to the current video;
Characteristic extracting module, for carrying out the selection of target frame to first frame image, to be set as feature templates, and to described current
The picture of each frame of video carries out target's feature-extraction;
Template renewal module for carrying out template matching to the picture after extraction, and updates template library according to matching result;
Image output module meets in advance for controlling the template library output matching result, and when determining the matching result
If when indicator conditions, exporting the corresponding video frame images of the matching result.
10. image recognition according to claim 9 tracks processing system, which is characterized in that the characteristic extracting module is also
For:
The picture is converted into grayscale image, and carries out Gradient Features extraction in the horizontal and vertical directions respectively;
The grayscale image is converted picture gradient map by the absolute value that the gradient value on each pixel is taken in the picture,
And gaussian filtering and normalized are sequentially carried out to the picture gradient map;
The Gradient Features of template image are extracted in the horizontal and vertical directions respectively, are taken in the template image each
The absolute value of gradient value on pixel;
Template gradient map is converted by the template image, and gaussian filtering process is carried out to the template gradient map;
The obtained template gradient map is carried out expanding calculation process, then is normalized and thresholding processing.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910774328.1A CN110472608A (en) | 2019-08-21 | 2019-08-21 | Image recognition tracking processing method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910774328.1A CN110472608A (en) | 2019-08-21 | 2019-08-21 | Image recognition tracking processing method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110472608A true CN110472608A (en) | 2019-11-19 |
Family
ID=68513262
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910774328.1A Pending CN110472608A (en) | 2019-08-21 | 2019-08-21 | Image recognition tracking processing method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110472608A (en) |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111611854A (en) * | 2020-04-16 | 2020-09-01 | 杭州电子科技大学 | Classroom condition evaluation method based on pattern recognition |
CN111738053A (en) * | 2020-04-15 | 2020-10-02 | 上海摩象网络科技有限公司 | Tracking object determination method and device and handheld camera |
CN112287906A (en) * | 2020-12-18 | 2021-01-29 | 中汽创智科技有限公司 | Template matching tracking method and system based on depth feature fusion |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6795567B1 (en) * | 1999-09-16 | 2004-09-21 | Hewlett-Packard Development Company, L.P. | Method for efficiently tracking object models in video sequences via dynamic ordering of features |
US8477998B1 (en) * | 2008-06-20 | 2013-07-02 | Google Inc. | Object tracking in video with visual constraints |
CN103324920A (en) * | 2013-06-27 | 2013-09-25 | 华南理工大学 | Method for automatically identifying vehicle type based on vehicle frontal image and template matching |
CN104658011A (en) * | 2015-01-31 | 2015-05-27 | 北京理工大学 | Intelligent transportation moving object detection tracking method |
CN107798329A (en) * | 2017-10-29 | 2018-03-13 | 北京工业大学 | Adaptive particle filter method for tracking target based on CNN |
CN108846854A (en) * | 2018-05-07 | 2018-11-20 | 中国科学院声学研究所 | A kind of wireless vehicle tracking based on motion prediction and multiple features fusion |
CN110084829A (en) * | 2019-03-12 | 2019-08-02 | 上海阅面网络科技有限公司 | Method for tracking target, device, electronic equipment and computer readable storage medium |
-
2019
- 2019-08-21 CN CN201910774328.1A patent/CN110472608A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6795567B1 (en) * | 1999-09-16 | 2004-09-21 | Hewlett-Packard Development Company, L.P. | Method for efficiently tracking object models in video sequences via dynamic ordering of features |
US8477998B1 (en) * | 2008-06-20 | 2013-07-02 | Google Inc. | Object tracking in video with visual constraints |
CN103324920A (en) * | 2013-06-27 | 2013-09-25 | 华南理工大学 | Method for automatically identifying vehicle type based on vehicle frontal image and template matching |
CN104658011A (en) * | 2015-01-31 | 2015-05-27 | 北京理工大学 | Intelligent transportation moving object detection tracking method |
CN107798329A (en) * | 2017-10-29 | 2018-03-13 | 北京工业大学 | Adaptive particle filter method for tracking target based on CNN |
CN108846854A (en) * | 2018-05-07 | 2018-11-20 | 中国科学院声学研究所 | A kind of wireless vehicle tracking based on motion prediction and multiple features fusion |
CN110084829A (en) * | 2019-03-12 | 2019-08-02 | 上海阅面网络科技有限公司 | Method for tracking target, device, electronic equipment and computer readable storage medium |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111738053A (en) * | 2020-04-15 | 2020-10-02 | 上海摩象网络科技有限公司 | Tracking object determination method and device and handheld camera |
WO2021208253A1 (en) * | 2020-04-15 | 2021-10-21 | 上海摩象网络科技有限公司 | Tracking object determination method and device, and handheld camera |
CN111738053B (en) * | 2020-04-15 | 2022-04-01 | 上海摩象网络科技有限公司 | Tracking object determination method and device and handheld camera |
CN111611854A (en) * | 2020-04-16 | 2020-09-01 | 杭州电子科技大学 | Classroom condition evaluation method based on pattern recognition |
CN111611854B (en) * | 2020-04-16 | 2023-09-01 | 杭州电子科技大学 | Classroom condition evaluation method based on pattern recognition |
CN112287906A (en) * | 2020-12-18 | 2021-01-29 | 中汽创智科技有限公司 | Template matching tracking method and system based on depth feature fusion |
CN112287906B (en) * | 2020-12-18 | 2021-04-09 | 中汽创智科技有限公司 | Template matching tracking method and system based on depth feature fusion |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111709311B (en) | Pedestrian re-identification method based on multi-scale convolution feature fusion | |
CN111652317B (en) | Super-parameter image segmentation method based on Bayes deep learning | |
CN112668483B (en) | Single-target person tracking method integrating pedestrian re-identification and face detection | |
CN113763424B (en) | Real-time intelligent target detection method and system based on embedded platform | |
CN110472608A (en) | Image recognition tracking processing method and system | |
CN105893946A (en) | Front face image detection method | |
CN107909081A (en) | The quick obtaining and quick calibrating method of image data set in a kind of deep learning | |
CN114758288A (en) | Power distribution network engineering safety control detection method and device | |
CN111931654A (en) | Intelligent monitoring method, system and device for personnel tracking | |
CN107316029A (en) | A kind of live body verification method and equipment | |
CN110991397B (en) | Travel direction determining method and related equipment | |
CN112434599A (en) | Pedestrian re-identification method based on random shielding recovery of noise channel | |
CN114049581A (en) | Weak supervision behavior positioning method and device based on action fragment sequencing | |
CN112686122B (en) | Human body and shadow detection method and device, electronic equipment and storage medium | |
CN111881775B (en) | Real-time face recognition method and device | |
CN109493370A (en) | A kind of method for tracking target based on spatial offset study | |
CN114708645A (en) | Object identification device and object identification method | |
Harish et al. | New features for webcam proctoring using python and opencv | |
CN116229052A (en) | Method for detecting state change of substation equipment based on twin network | |
CN110197123A (en) | A kind of human posture recognition method based on Mask R-CNN | |
CN112348011B (en) | Vehicle damage assessment method and device and storage medium | |
CN113920168A (en) | Image tracking method in audio and video control equipment | |
CN111310679A (en) | Goose walking gait feature recognition method based on machine vision | |
CN113255549A (en) | Intelligent recognition method and system for pennisseum hunting behavior state | |
CN117011335B (en) | Multi-target tracking method and system based on self-adaptive double decoders |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20191119 |
|
RJ01 | Rejection of invention patent application after publication |