CN110188693A - Improved complex environment vehicle characteristics extract and parking method of discrimination - Google Patents
Improved complex environment vehicle characteristics extract and parking method of discrimination Download PDFInfo
- Publication number
- CN110188693A CN110188693A CN201910464427.XA CN201910464427A CN110188693A CN 110188693 A CN110188693 A CN 110188693A CN 201910464427 A CN201910464427 A CN 201910464427A CN 110188693 A CN110188693 A CN 110188693A
- Authority
- CN
- China
- Prior art keywords
- feature
- harr
- highway
- discrimination
- vehicle
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/254—Fusion techniques of classification results, e.g. of results related to same input data
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/20—Image preprocessing
- G06V10/25—Determination of region of interest [ROI] or a volume of interest [VOI]
-
- 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
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/50—Context or environment of the image
- G06V20/52—Surveillance or monitoring of activities, e.g. for recognising suspicious objects
- G06V20/54—Surveillance or monitoring of activities, e.g. for recognising suspicious objects of traffic, e.g. cars on the road, trains or boats
-
- G—PHYSICS
- G08—SIGNALLING
- G08G—TRAFFIC CONTROL SYSTEMS
- G08G1/00—Traffic control systems for road vehicles
- G08G1/01—Detecting movement of traffic to be counted or controlled
- G08G1/017—Detecting movement of traffic to be counted or controlled identifying vehicles
- G08G1/0175—Detecting movement of traffic to be counted or controlled identifying vehicles by photographing vehicles, e.g. when violating traffic rules
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/25—Fusion techniques
- G06F18/259—Fusion by voting
-
- 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
- G06V20/44—Event detection
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Multimedia (AREA)
- Data Mining & Analysis (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Image Analysis (AREA)
Abstract
Extract and stop the invention discloses a kind of improved complex environment vehicle characteristics method of discrimination, from highway actual environment, for this special scenes of highway, after analyzing highway exception parking test problems and the difficult point based on video, primary study highway complex condition vehicle Haar-like feature extraction, and Adaboost cascade classifier parking discrimination model be applied to highway scene when deficiency, improvement is made that the highway Parking detection algorithm based on Haar-like+Adaboost, improve the performance of highway Parking detection algorithm, it can reduce in illumination variation, difference in the class of vehicle Harr-like feature when visual angle change and dimensional variation, improve feature learning efficiency, it improves high The detection accuracy and detection system generalization ability of fast highway exception parking event.
Description
Technical field
The present invention relates to Vehicle Detection technology and data processing field, in particular to a kind of improved complex environment vehicle
Stop method of discrimination under Harr-like feature extraction and Adaboost frame.
Background technique
There are great security risk, real-time Parking state differentiations to mention in real time for highway exception parking
For the traffic information of express highway section, automation services preferably are provided for traffic administration personnel.Study complicated open-air atmosphere
Lower vehicle Harr-like feature extraction and existing traffic events can effectively be avoided based on Adaboostd feature learning optimization method
Detection system there is a situation where complex environment variation when occur as soon as high erroneous detection, missing inspection, be conducive to promote entire traffic events
The robustness and robustness of detection system.
Existing patent and paper are read, in terms of the vehicle on highway identification based on video, with digitized map in recent years
As the rapid development of processing technique, differentiate that effect is also more significant.It is broadly divided into four classes:
The first kind is based drive vehicle checking method, is mainly this feature of movement, analysis view using target
Changing unit in frequency between successive frame, and detect vehicle.Currently based on movement vehicle checking method there are mainly three types of: light
Stream method, frame difference method and Background difference.Second class is the vehicle checking method based on priori knowledge, the priori of application target, scene
Knowledge positions vehicle from image, these priori knowledges mainly include symmetry, color, shade, geometrical characteristic, texture and vehicle
Lamp.Third class is 2D the or 3D model for constructing vehicle in advance using priori knowledge based on the vehicle checking method based on template, is used
By model projection to image space, model projection is matched with characteristics of image for suitable transformation, takes best match as detection
As a result.4th class is the vehicle checking method based on feature learning, and two classification that problem is considered as a vehicle and non-vehicle are asked
Topic, comprising finding the most suitable line of demarcation that can distinguish two classes in this identifying system, however since illumination becomes
Change, the presence of dimensional variation and visual angle change, Accurate classification is not an easy thing.Currently, being based on Harr-like+
The algorithm of target detection of Adaboost has preferable target classification ability, and required image pattern is less, and feature training is not easy
Over-fitting is generated, is suitable for solving the problems, such as complex scene vehicle detection.
Summary of the invention
In view of this, an object of the present invention is to provide, a kind of improved complex environment vehicle characteristics are extracted and parking is sentenced
Other method.The present invention is from highway actual environment, for difference mistake in complex condition vehicle Harr-like feature class
Big problem is proposed and is stopped under vehicle Harr-like feature extraction and Adaboost frame under a kind of improved complex environment
Event method of discrimination, can quickly and to effectively improve algorithm be visual angle change to highway, illumination variation and dimensional variation
Adaptability improves the generalization ability of Parking accuracy in detection and detection system.The second object of the present invention is to provide one kind
System.
The purpose of the present invention is what is be achieved through the following technical solutions:
Improved complex environment vehicle characteristics of the invention extract and parking method of discrimination, it is characterised in that: including following
Step:
Step S1: visual angle amendment is carried out to sequence of video images;
Step S2: illumination amendment is carried out;
Step S3: potential target region candidate frame extracts;
Step S4: scale amendment is carried out;
Step S5: event differentiation is carried out by Harr-like statistical learning classifier.
Particularly, the step S1 includes following sub-step:
Step S11: region of interesting extraction is carried out;
Step S12: zone viewing angle differentiation is carried out;
Step S13: Image Reversal is carried out.
Particularly, the step 2 includes following sub-step:
Step S21: single Gaussian Background modeling;
Step S22:ROI regional luminance differentiates;
Step S23:Gamma transformation.
Particularly, the step S3 includes following sub-step:
Step S31: using the mixed Gaussian background modeling of low context update rate;
Step S32: object discrimination is stood;
Step S33: prospect two-value picture dilation erosion;
Step S34: it is excluded based on the local hot spot that Scanny is extracted;
Step S35: output potential target candidate region.
Particularly, the step S4 includes following sub-step:
Step S41: load Harr-Like feature Weak Classifier;
Step S42: potential target region size-varied window traversal;
Step S43: detection window feature remaps.
Particularly, the step S5 includes following sub-step:
Step S51: each Weak Classifier Harr-like characteristic value size is exported;
Step S52: Weak Classifier Nearest Neighbor with Weighted Voting is used;
Step S53: output test result.
Particularly, the method is carried out using the cooperation of improved Adaboost feature learning system, comprising:
Early period rejects atypical characteristics, in the training process of each Weak Classifier, to all Harr-like features
Whether before traversal screening, prejudging first is atypical characteristics, if it is, not carrying out operation to the feature of the type;
For partial points bulk feature existing for vehicle under highway scene, the center contour feature of mutative scale is designed,
It is allowed to participate in during Adaboost feature learning, classification capacity is obtained in the center contour feature most preferably by algorithm
, improve classifying quality.
The second object of the present invention is to what is be achieved through the following technical solutions, this kind of complex environment vehicle characteristics extract and stop
Vehicle judgement system, comprising:
Visual angle correction module: the image of different scenes is carried out to the Harr- that visual angle is unified, caused by diminution visual angle is different
Difference in like feature class;
Illumination correction module: by detecting real-time luminosity, brightness regulation is carried out to image, caused by reducing illumination variation
Difference in Harr-like feature class;
Potential target region candidate frame extraction module: foreground target candidate region and target scale are got by background subtraction
Information reduces unnecessary calculation resources waste when sliding detection window traversal, improves detection real-time, while being next step ruler
Degree amendment provides candidate region information;
Scale correction module: carrying out mutative scale detection window traversal to candidate region, loads good trained classification
Device file carries out Harr-like feature in the feature of detection window to the Harr-like feature of wherein each Weak Classifier
It remaps, corresponding Harr-like characteristic value is calculated;
Harr-like characteristic statistics learn discrimination module: to multiple distinctive Harr- of Weak Classifier obtained in the previous step
Like characteristic value is differentiated, obtains final judging result by Nearest Neighbor with Weighted Voting.
The beneficial effects of the present invention are: the present invention, from highway actual environment, for highway, this is specific
Scene, after analyze highway exception parking test problems and the difficult point based on video, primary study highway is multiple
It is public to be applied to high speed for vehicle Haar-like feature extraction and Adaboost cascade classifier parking discrimination model under the conditions of miscellaneous
The deficiency when scene of road is made that improvement to the highway Parking detection algorithm based on Haar-like+Adaboost,
The performance of highway Parking detection algorithm is improved, can reduce in illumination variation, visual angle change and dimensional variation
When vehicle Harr-like feature class in difference, improve feature learning efficiency, improve highway exception parking event detection
Precision and detection system generalization ability.
Other advantages, target and feature of the invention will be illustrated in the following description to a certain extent, and
And to a certain extent, based on will be apparent to those skilled in the art to investigating hereafter, Huo Zheke
To be instructed from the practice of the present invention.Target and other advantages of the invention can be wanted by following specification and right
Book is sought to be achieved and obtained.
Detailed description of the invention
To make the objectives, technical solutions, and advantages of the present invention clearer, below in conjunction with attached drawing to the present invention make into
The detailed description of one step, in which:
Fig. 1 is that Harr-like and Adaboost detects overview flow chart;
Fig. 2 is improved Adaboost feature learning algorithm flow chart
Fig. 3 is the vehicle mutative scale central feature template for designing addition herein.
Fig. 4 is that mutative scale detection window feature remaps schematic diagram;
Fig. 5 is atypia Harr-like feature example schematic;
Fig. 6 is five kinds of Harr-like feature templates schematic diagrames;
Fig. 7 is center characteristic type schematic diagram;
Fig. 8 is the nine kinds of center Harr-like profile feature templates and its characterization schematic diagram to vehicle local message;
Fig. 9 is the positive and negative sample value sorted lists of specific Haar-like feature.
Specific embodiment
Hereinafter reference will be made to the drawings, and a preferred embodiment of the present invention will be described in detail.Obviously, described embodiment
Only a part of the embodiment of the present invention, instead of all the embodiments.Based on the embodiments of the present invention, the common skill in this field
Art personnel every other embodiment obtained without making creative work belongs to the model that the present invention protects
It encloses.
As depicted in figs. 1 and 2, improved complex environment vehicle characteristics of the invention extraction and parking method of discrimination, including
Following steps:
Step S1: visual angle amendment is carried out to sequence of video images;
Step S2: illumination amendment is carried out;
Step S3: potential target region candidate frame extracts;
Step S4: scale amendment is carried out;
Step S5: event differentiation is carried out by Harr-like statistical learning classifier.
This method hints obliquely at transformation first against visual angle change, by area-of-interest visual angle automatic identification and two dimensional image,
Image potential target region is overturn, the significant changes of vehicle appearance caused by different perspectives are reduced;For illumination variation,
The background picture as obtained by the modeling of single Gaussian Background carries out real-time luminosity detection, is then converted by Gamma so that vehicle brightness
Keep small range fluctuation;For dimensional variation, the process that original Harr-like Feature prototype is mapped in size-varied window
It is made that improvement.
Secondly for the water stain local hot spot interference waited under highways scene of tree shade cloud shadow, a kind of Adaboost frame is studied
The candidate region extraction scheme for updating background model under frame based on lag, passes through comparison current image and background picture scanny line
Information is managed, candidate region is improved and extracts accuracy.And Harr-like characteristic statistics learning algorithm is improved, it rejects non-
Typical Harr-like feature improves training speed under the premise of guaranteeing training effect.Finally, being directed to highway scene
The characteristics of point block distribution is presented in lower vehicle local feature, and design is added to mutative scale Harr-like on the basis of original algorithm
Central feature is allowed to participate in during the Feature Selection based on Adaboost, improve to vehicle target and non-vehicle target
Classification capacity.
Parking differentiation side under improved complex environment vehicle Harr-like feature extraction of the invention and Adaboost frame
Method mainly comprises the following modules:
(1) visual angle correction module: for the image of different scenes to be carried out visual angle unification, caused by reducing visual angle difference
Difference in Harr-like feature class, specifically includes:
Firstly, carrying out region of interesting extraction: the extraction of area-of-interest belongs to the modified pre-processing in visual angle, main
Purpose is that lane region is isolated from image, is gone unless the influence that car light is extracted in lane region, and obtain visual angle amendment area
The original coordinates in domain.By analyzing video image under highway scene, in the present embodiment, area-of-interest follow with
Lower principle extracts:
1. removing non-rice habitats region.Since vehicle is only possible to traveling in road area, rather than trees existing for road area
Etc. targets not within the vehicle characteristics extraction scope of this paper.
2. removing road remote areas.There is largely blocking between the distant place of road, each vehicle, and target texture is special
It goes on a punitive expedition in smudgy, so that the extraction interference to vehicle characteristics is larger, in order to guarantee accuracy that vehicle characteristics extract, herein
Biggish region is interfered to be not counted in area-of-interest at a distance in road.
3. removing the interference of character present in image.Due to containing road in the collected video image of part camera
The characters such as information, the upper left corner and the lower right corner of image show the character of road name and time, and also have on the road in left side
The white characters of video camera information, extraction and study to vehicle characteristics have certain influence, therefore select in area-of-interest
It should give and remove when taking.In conclusion can not only reduce vehicle characteristics extraction to the Rational choice of interesting image regions
Interference provides reference information for image aspects correction, while also improving the processing speed of algorithm.
Finally, carrying out visual angle differentiation according to interested area information.It, can according to the four of quadrangle area-of-interest angle points
To determine that the scene camera is to be located at left side or right side to the shooting visual angle of vehicle, is overturn, can be made by image level
The vehicle image that takes of all scene cameras is the same side visual angle (is left side perspective or be right side perspective).
(2) illumination correction module: by detecting real-time luminosity, brightness regulation is carried out to image, illumination variation is reduced and causes
Harr-like feature class in difference;
When area-of-interest (ROI) illumination is more abundant, image grayscale distribution concentrates on high bright part, this leads to figure
As whole whiting, vehicle characteristics are unobvious;And when area-of-interest (ROI) illumination is very insufficient, image grayscale distribution
Shadow portion is concentrated on, vehicle characteristics are fuzzy.Therefore, there can be the finger of an automatic discrimination to the intensity of illumination of ROI region
Mark, to make adaptive adjustment to image grayscale distribution using this index, so that vehicle image feature is either gone back bright
It is the conspicuousness for guaranteeing minutia under dim scene, has great importance.
Present invention uses a kind of, and the brightness based on model of place differentiates scheme.It obtains and is continuously regarded in a period of time first
Frequency image sequence, the data source as scene model foundation in this time.Then according to the Gaussian Background of Fast Learning rate
Modeling method obtains a background picture of current period.Image grayscale statistics is carried out to the picture, is calculated in area-of-interest
The average value of pixel brightness contribution obtains ROI region brightness discriminant index.
In the case where obtaining current scene after background picture, pixel intensity α under current scene can be calculated, the figure used herein
Image brightness index is by calculating the ratio of picture area-of-interest average gray value under current background picture and normal illumination come table
Show.When α is greater than 1, current scene is partially bright, and when α is less than 1, current scene is partially dark.Wherein K indicates to feel under normal illumination environment
Interest region average gray value, the value are adjustable threshold, according to the appropriate adjustment of actually detected progress.Wherein
(n indicates the number of area-of-interest pixel, xiIndicate the gray value of area-of-interest pixel.)
Brightness correction is carried out to picture in time according to the luminance information under current scene, it is ensured that the image local of vehicle
Feature will not be too strong due to illumination and excessively weak and becomes to be not easy to extract.Most common illumination modification method is exactly the school Gamma
Just.So-called Gamma correction is exactly to edit to the gamma curve of image, with to image carry out non-linear tone editor method,
The dark parts and light-colored part in picture signal are detected, and increase the two ratio, to improve picture contrast effect.It is logical
In normal situation, Gamma correction takes fixed correction value, but under highway scene, illumination is sometimes bright sometimes dark, at this time if
Gamma correction cannot adaptively be adjusted with the dynamic change of illumination, then be likely to occur the effect run counter to desire.The present invention mentions
One kind is gone out and amendment scheme is shone based on the adaptive optical that area-of-interest (ROI) brightness differentiates.Specific amendment scheme is as follows:
Wherein I (x, y) indicates current image in the pixel value size that coordinate is at (x, y) point, and wherein Gamma converts index
γ=k* α, i.e., when brightness of image is higher, Gamma index is greater than 1, and high luminance value regions are improved after transformation
Contrast;When brightness of image is partially obscure, Gamma index improves the comparison of low brightness area less than 1 after transformation
Degree.Specific step is as follows: assuming that there is a pixel in image, value is 200, then this pixel is corrected have to carry out as
Lower step:
1. normalization: pixel value is converted to the real number between 0~1.Algorithm is as follows: (i+0.5)/256 includes 1 here
Division and 1 add operation.For pixel A, corresponding normalized value is 0.783203.
2. precompensation: according to formula, the data after finding out pixel normalization are using 1/gamma as the respective value of index.This
Step seeks exponent arithmetic comprising one.If gamma value is 2.2,1/gamma 0.454545, the A value after normalization is carried out pre-
The result of compensation is exactly 0.783203^0.454545=0.894872.
3. renormalization: the integer value between 0~255 will be changed to by the real number value contravariant of precompensation.Specific algorithm are as follows:
This step of f*256-0.5 includes a multiplication and a subtraction.Continuous precedent, the precompensation result 0.894872 of A is substituted into
Above formula, corresponding pixel value is 228 after obtaining A precompensation, this 228 be exactly last feeding display data.
As described above if directly pressing formula programming, it is assumed that the resolution ratio of image is 800*600, is carried out to it
Gamma correction needs to be implemented 480,000 floating number multiplication, division and exponent arithmetic.Efficiency is too low, and real-time effect is not achieved at all
Fruit.
For above situation, a kind of fast algorithm is proposed, if it is possible to the pixel value range for knowing image, for example, 0
Integer between~255, then any one pixel value can only be some in 0 to 255 this 256 integers in image;?
In situation known to gamma value, any integer between 0~255, after " normalization, precompensation, renormalization " operation,
It is corresponding the result is that unique, and also fall in 0~255 within the scope of this.
As in the previous example, it is known that gamma value is 2.2, and the original value of pixel A is 200, so that it may it is corresponding to acquire the A after gamma correction
Pre-compensation value be 228.Based on the above principles, we need to execute primary precompensation behaviour only for each integer between 0~255
Make, its corresponding pre-compensation value is stored in a gamma correction look-up table (LUT:Look Up Table) pre-established, just
The table can be used, Gamma correction is carried out to image of any pixel value between 0~255.
In conclusion may be implemented fast by good Gamma correction look-up table (LUT:Look Up Table) of off-line calculation
Fast mapping transformation ensure that the real-time of algorithm.Generally, enable vehicle characteristics that there is robustness to light change, it is real
Image irradiation adaptive correction is showed.
(3) foreground target candidate region and target potential target region candidate frame extraction module: are got by background subtraction
Dimensional information reduces unnecessary calculation resources waste when sliding detection window traversal, improves detection real-time, while being next
It walks scale amendment and candidate region information is provided;
Foreground target region is got by background subtraction first, then background picture and current image corresponding region are carried out
Scanny texture blending.Under highway scene, pavement texture feature does not occur significantly to become with the variation of illumination
Change, once opposite vehicle is parked in after road area, significant change can occur for the textural characteristics in the region.According to this original
Reason illustrates it is that light luminance becomes if the textural characteristics of target area background picture and the textural characteristics of real time picture are close
Change and result in the noise that Gauss extracts foreground target, which is determined as no vehicle target.By this method, it significantly mentions
The high extraction accuracy of vehicle candidate region.
As shown in figure 3, the Scanny texture in the region same region of Scanny texture and background has in vehicle target region
There is marked difference;And the region for there is local hot spot, for the Scanny texture of corresponding region, background picture and to be checked
Mapping piece is without significant difference.Based on this, then local hot spot interference can be excluded, candidate region is improved and extracts accuracy.
(4) scale correction module: mutative scale detection window traversal is carried out to candidate region, load is trained point good
Class device file carries out Harr-like feature in the spy of detection window to the Harr-like feature of wherein each Weak Classifier
Sign remaps, and corresponding Harr-like characteristic value is calculated;
Existing vehicle Harr-like characteristic extraction procedure, using the size of fixed proportion and the length-width ratio of fixed proportion
The mapping of Feature prototype to detection window is carried out, this method is not able to satisfy vehicle target scale under highway scene and changes model
Enclose this larger actual conditions.
After the potential target region to different scale carries out adaptive-filtering, replace original ruler this paper presents a kind of
Degree is fixed, and the fixed detection window feature extracting method of length-width ratio is not fixed, the unfixed potential target area of length-width ratio with scale
Domain is detection window, is extracted after remapping to feature.The method can reduce training sample and actually detected target it
Between feature Distribution value gap, effectively increase detection accuracy.The specific method of mutative scale Feature Mapping is as shown in Figure 4:
If the size of training sample picture is 24*24 pixel;The size of potential target region detection window traversal somewhither
For W*H;A certain Harr-like edge feature prototype is Feature (x1,y1,x2,y2, s=1, t=2), i.e. this feature rectangle exists
Position in the samples pictures of 24*24 size are as follows: top left co-ordinate is (x1,y1), bottom right angular coordinate is (x2,y2), and by one black one
White two congruent rectangles are constituted.
Feature Feature (x1,y1,x2,y2, s=1, t=2) and the detection of W*H size is mapped to by the picture of 24*24 size
Window becomes Feature (x '1,y’1,x’2,y’2, s=1, t=2) algorithm it is as follows:
1. calculating scale factor:
X-direction scale factor α=W/24;Y-direction scale factor β=H/24.
2. calculating upper left corner mapping point (x '1,y’1):
x’1=[α * x1], y '1=[β * y1],
Wherein [] indicates to be rounded inner element downwards.
3. calculating lower right corner mapping point (x '2,y’2):
In order to guarantee that the rectangle after remapping still meets (s, t) condition of condition rectangle, converted by scale factor
Coordinate is adjusted after ratio.
4. obtain remapping by feature detection window characteristic information (the characteristic rectangle upper left corner and lower right position and
Feature templates type) Feature (x '1,y’1,x’2,y’2, s=1, t=2), it is counted in advance by potential target candidate region later
Good integral picture calculates the characteristic value size of the Harr-like feature.
(5) Harr-like characteristic statistics learn discrimination module: distinctive to multiple Weak Classifiers obtained in the previous step
Harr-like characteristic value is differentiated, obtains final judging result by Nearest Neighbor with Weighted Voting.
T best Weak Classifier h are obtained by training1(x),...hT(x), it can be combined into one in the following way
Strong classifier:
Wherein
When the strong classifier treats a width image to be detected, it is equivalent to and all Weak Classifiers is allowed to vote, then to voting results
According to error rates of weak classifiers weighted sum, the result of voting weighted summation is obtained compared with average voting results final
As a result.In practical applications, appropriate adjustment can be carried out to the threshold value of Nearest Neighbor with Weighted Voting, to reach required best classification effect
Fruit.
The detailed process introduced below for carrying out statistical learning to vehicle Harr-like feature by Adaboost algorithm:
Vehicle Haar-like characteristic statistics based on Adaboost learn rudimentary algorithm main flow are as follows:
Firstly, as follows to the algorithm description of Adaboost training strong classifier:
1) a series of training sample that highway scene history video acquisitions arrive is given, wherein indicating it for negative sample
(non-vehicle samples pictures) indicate it for positive sample (vehicle samples pictures).N is training sample total number;
2) weight w is initialized1,i=D (i);Wherein(to negative sample) or(to positive sample), m and
L is respectively the quantity of negative sample (non-vehicle target) and positive sample (vehicle target).
3) to t=1 ... T:
A. normalized weight, so that the sum of weight of all positive negative samples is 1:
A. to each feature f, one Weak Classifier h (x, f, p, θ) of training;Calculate the Weak Classifier of corresponding all features
Weight qtError rate (εf):
εf=∑iqi|h(xi,f,p,θ)-yi|
B. optimal Weak Classifier h is chosent(x), which possesses minimum weight error rate εt:
εt=minf,p,θ∑iqi|h(xi,f,p,θ)-yi|=∑iqi|h(xi,ft,pt,θt)-yi|
ht(x)=h (x, ft,pt,θt)
C. according to this best Weak Classifier, weight is adjusted:
Wherein ei=0 i-th of sample of expression is correctly classified, ei=1 indicates that i-th of sample is classified by mistake;
4) last strong classifier are as follows:
Wherein:
Wherein for λ in 0 to 1 range, general value is 0.5, the parameter can be adjusted in practical applications with full
Sufficient actual demand.If increasing the value, the false detection rate of strong classifier is reduced, but leak detection can increase;If reducing the value,
The false detection rate of strong classifier increases, but leak detection can be reduced.
Secondly, being described below to Weak Classifier model:
One complete vehicle target classification Weak Classifier h (x, f, p, θ) is made of following three elements:
1) it under current positive and negative sample weights, is obtained by traversing all samples pictures and all Haar-like characteristic types
That the smallest feature of classification weighting fault rate arrived;
2) by traversal and the characteristic value of the f feature for all positive and negative samples pictures that sort, the characteristic value of optimal classification is realized
Boundary θ;
3) p in sign of inequality direction, if the calculated result of clarification of objective f is greater than above-mentioned threshold θ and will be judged as just
Sample, then p takes just;Otherwise p takes negative sign.
To sum up, the construction of Weak Classifier is as follows:
The step of being trained to Weak Classifier is as follows:
One Weak Classifier (feature Feature (x of training1,y1,x2,y2, s, t)) it is exactly the case where present weight is distributed
Under, determine Feature (x1,y1,x2,y2, s, t) optimal threshold so that this Weak Classifier (feature Feature (x1,y1,
x2,y2, s, t)) it is minimum to the weighting classification error of all training samples.Choosing a best Weak Classifier is exactly to select that right
The error in classification of all positive negative training samples that Weak Classifier minimum in all Weak Classifiers, i.e. selection Feature
(x1,y1,x2,y2, s, t) and be positive one of the fixed reference feature that negative sample detects.
For each feature Feature (xi,yi,xj,yj, s, t), the characteristic value of all positive negative training samples is calculated, and will
It sorts.By scanning a sorted characteristic value, an optimal threshold value can be determined for this feature, to be trained to
One Weak Classifier.Specifically, to each element in sorted table, following four value is calculated:
1) weight of rolling stock sample and T+;
2) weight and T of whole non-face samples-;
3) the sum of weight of vehicle sample before this element S+;
4) the sum of weight of non-vehicle sample after this element S-;
When the characteristic value for choosing currentElementWith a characteristic value before itBetween number as threshold value when, institute
Obtained Weak Classifier just separates sample at currentElement --- and that is the corresponding Weak Classifier of this threshold value will work as
All elements before preceding element are classified as vehicle (or non-vehicle), and all elements (contained) after currentElement are classified as non-vehicle
(or vehicle).Obtain the error in classification obtained under the threshold value are as follows:
E=min (S++(T--S-),S-+(T+-S+))
Scanning one time from the beginning to the end by the table scan that this is sorted can be to make error in classification for Weak Classifier selection
The smallest threshold value (optimal threshold), that is, have chosen a best Weak Classifier.Since in sorted lists, optimal threshold is only
The adjacent intersection of positive and negative sample elements is likely located at, therefore during realization to the algorithm herein, not to all
Element seeks error in classification, but seeks point to the adjacent node element of positive and negative sample is occurred in sorted element list
Class error, is screened later, improves the efficiency for seeking Weak Classifier optimal threshold θ in this way.It is as shown in Figure 9:
Characteristic value calculating is carried out to the target area extracted during atual detection, works as e=S++(T--S-) when,
It indicates to be determined as vehicle target if this feature value is greater than the optimal threshold;If this feature value is less than the optimal threshold,
It is determined as non-vehicle target.Work as e=S-+(T+-S+) when, it indicates to be determined as vehicle if this feature value is less than the optimal threshold
Target;If this feature value is greater than the optimal threshold, it is determined as non-vehicle target.
Again, the composition of strong classifier is as follows:
After T iteration, T best Weak Classifier h are obtained1(x),...hTIt (x), can group in the following way
Synthesize a strong classifier:
Wherein:
It when this strong classifier treats a width image to be detected, is equivalent to and all Weak Classifiers is allowed to vote, then ballot is tied
Fruit obtains compared with average voting results final according to error rates of weak classifiers weighted sum, by the result of voting weighted summation
Result.In practical applications, appropriate adjustment can be carried out to the threshold value of Nearest Neighbor with Weighted Voting, to reach required best classification effect
Fruit.
The present invention is matched using improved Adaboost feature learning system with Harr-like, is specifically included:
(1)) early period rejects part atypia Harr-like feature
It is too big with such feature calculation characteristic value randomness for occupying the rectangular characteristic of four pixels, by experiment
Verifying, can not search out suitable threshold value.This step, feature quantity can reduce 3,243;Again because of the vehicle characteristics of instruction collection
Middle part is all concentrated on, the rectangular characteristic on edge will not be especially big to the contribution for improving classifying quality, therefore can suitably reduce
The rectangular characteristic at edge, then the quantity of rectangular characteristic can be further reduced 652.As shown in Figure 5.
Therefore, it for the Adaboost integrated study of Harr-like feature under highway scene, can exclude small
The extraction and calculating of Harr-like feature and edge Harr-like feature improve training under the premise of guaranteeing training effect
Speed.It is somebody's turn to do in of the invention into scheme, defines small Harr-like feature first are as follows: Small_Feature (x1,y1,
x2,y2, s, t), and five parameters for describing this feature type should meet simultaneously following six condition:
(1. s, t) ∈ { (1,2), (2,1), (1,3), (3,1), (2,2) }
2.x1∈{1,2,...SCALE-s,SCALE-s+1}
3.y1∈{1,2,...SCALE-t,SCALE-t+1}
4.x2∈ X={ x1+s-1,x1+2*s-1,...,x1+(p-1)*s-1,x1+p*s-1}
5.y2∈ Y={ y1+t-1,y1+2*t-1,...,y1+(q-1)*t-1,y1+q*t-1}
6.(x2-x1+1)*(y2-y1+1)≤4
Wherein:
Secondly, defining edge Harr-like feature are as follows: Margin_Feature (x1,y1,x2,y2, s, t) and the spy is described
Five parameters of sign type should meet simultaneously following six condition:
(1. s, t) ∈ { (1,2), (2,1), (1,3), (3,1), (2,2) }
2.x1∈{1,2,...SCALE-s,SCALE-s+1}
3.y1∈{1,2,...SCALE-t,SCALE-t+1}
4.x2∈ X={ x1+s-1,x1+2*s-1,...,x1+(p-1)*s-1,x1+p*s-1}
5.y2∈ Y={ y1+t-1,y1+2*t-1,...,y1+(q-1)*t-1,y1+q*t-1}
6.(x1=1 ∩ x2=1) ∪ (y1=1 ∩ y2=1) ∪ (x1=SCALE ∩ x2=SCALE) ∪ (y1=SCALE ∩ y2
=SCALE)
Wherein:
(2) design is suitable for the Harr-like feature templates of highway scene vehicle characteristics description
Firstly, the size of Haar-like character value reflects the grey scale change situation of image.At present with often used
Five Harr-like feature templates and corresponding type specification it is as shown in Figure 6.These features preferable can must characterize vehicle
In Edge texture feature vertically and horizontally:
But under highway scene, vehicle target exists simultaneously the apparent central feature (left and right two of such as vehicle
The vehicle window of side), vehicle target and non-vehicle target can be distinguished well.Therefore herein in original Harr-like
In feature templates library, design is added to central feature template.The center die plates are defined as follows, i.e., white area picture in rectangle D
The sum of element subtract black region pixel value and, as the characteristic value of the central feature, as shown in Figure 7:
Feature(x1,y1,x2,y2,s,t)
=intgePicture (1)+intgePicture (4)-intgePicture (2)-intgePicture (3)+
(intgePicture(5)+intgePicture(8)-intgePicture(6)-intgePicture(7))*2
Wherein, square of the integrogram that intgePicture (x) expression former potential target region picture is calculated at point x
Battle array calculated value.
As shown in figure 8, the central feature that the present invention designs is segmented altogether several types shown in figure.
Present design, fully considered highway changeable scene downwards angle of visibility variation, illumination variation and dimensional variation because
Element has improved corresponding Harr-like characteristic extraction procedure for every kind of influence factor that may be present.Generally improve
The robustness and accuracy of detection system.
The present invention is being analyzed for this special scenes of highway based on view from highway actual environment
After the highway exception parking test problems and difficult point of frequency, primary study highway complex condition vehicle Haar-
Like feature extraction and Adaboost cascade classifier parking discrimination model are applied to deficiency when highway scene, right
Highway Parking detection algorithm based on Haar-like+Adaboost is made that improvement, finally improves highway
The performance of Parking detection algorithm.It specifically includes that
1. in terms of complex condition vehicle Harr-like feature extraction, being directed to visual angle change respectively, passing through area-of-interest
Visual angle automatic identification and two dimensional image hint obliquely at transformation, overturn to image potential target region, reduce caused by different perspectives
The significant changes of vehicle appearance;For illumination variation, the background picture as obtained by the modeling of single Gaussian Background carries out real-time luminosity inspection
It surveys, is then converted by Gamma so that vehicle brightness keeps small range fluctuation;For dimensional variation, to original Harr-like
The process that Feature prototype maps in size-varied window is made that improvement.The experimental results showed that this series of improvement effectively subtracts
Small vehicle Haar-like feature difference in the class under changeable scene, generalization ability and the identification for improving detection algorithm are quasi-
True property.
2. cascade classifier discrimination model improves aspect under Adaboost frame, high speed is waited first against tree shade cloud shadow is water stain
Local hot spot interference under highway scene, studies the candidate region for updating background model under a kind of Adaboost frame based on lag
Extraction scheme improves candidate region and extracts accuracy by comparison current image and background picture scanny texture information.And
Harr-like characteristic statistics learning algorithm is improved, atypia Harr-like feature is rejected, is guaranteeing training effect
Under the premise of, improve training speed.Finally, the spy of point block distribution is presented for vehicle local feature under highway scene
Point, design is added to mutative scale Harr-like central feature on the basis of original algorithm, is allowed to participate in based on Adaboost's
During Feature Selection, the classification capacity to vehicle target and non-vehicle target is improved.
Verified, system proposed in this paper can improve detection accuracy under the premise of guaranteeing event detection real-time.
It should be appreciated that the embodiment of the present invention can be by computer hardware, the combination of hardware and software or by depositing
The computer instruction in non-transitory computer-readable memory is stored up to be effected or carried out.Standard volume can be used in the method
Journey technology-includes that the non-transitory computer-readable storage media configured with computer program is realized in computer program,
In configured in this way storage medium computer is operated in a manner of specific and is predefined --- according in a particular embodiment
The method and attached drawing of description.Each program can with the programming language of level process or object-oriented come realize with department of computer science
System communication.However, if desired, the program can be realized with compilation or machine language.Under any circumstance, which can be volume
The language translated or explained.In addition, the program can be run on the specific integrated circuit of programming for this purpose.
In addition, the operation of process described herein can be performed in any suitable order, unless herein in addition instruction or
Otherwise significantly with contradicted by context.Process described herein (or modification and/or combination thereof) can be held being configured with
It executes, and is can be used as jointly on the one or more processors under the control of one or more computer systems of row instruction
The code (for example, executable instruction, one or more computer program or one or more application) of execution, by hardware or its group
It closes to realize.The computer program includes the multiple instruction that can be performed by one or more processors.
Further, the method can be realized in being operably coupled to suitable any kind of computing platform, wrap
Include but be not limited to PC, mini-computer, main frame, work station, network or distributed computing environment, individual or integrated
Computer platform or communicated with charged particle tool or other imaging devices etc..Each aspect of the present invention can be to deposit
The machine readable code on non-transitory storage medium or equipment is stored up to realize no matter be moveable or be integrated to calculating
Platform, such as hard disk, optical reading and/or write-in storage medium, RAM, ROM, so that it can be read by programmable calculator, when
Storage medium or equipment can be used for configuration and operation computer to execute process described herein when being read by computer.This
Outside, machine readable code, or part thereof can be transmitted by wired or wireless network.When such media include combining microprocessor
Or other data processors realize steps described above instruction or program when, invention as described herein including these and other not
The non-transitory computer-readable storage media of same type.When the website according to the present invention based on big data log analysis
When intrusion detection method and technology program, the invention also includes computers itself.
Computer program can be applied to input data to execute function as described herein, to convert input data with life
At storing to the output data of nonvolatile memory.Output information can also be applied to one or more output equipments as shown
Device.In the preferred embodiment of the invention, the data of conversion indicate physics and tangible object, including the object generated on display
Reason and the particular visual of physical objects are described.
Finally, it is stated that the above examples are only used to illustrate the technical scheme of the present invention and are not limiting, although referring to compared with
Good embodiment describes the invention in detail, those skilled in the art should understand that, it can be to skill of the invention
Art scheme is modified or replaced equivalently, and without departing from the objective and range of the technical program, should all be covered in the present invention
Scope of the claims in.
Claims (8)
1. improved complex environment vehicle characteristics extract and parking method of discrimination, it is characterised in that: the following steps are included:
Step S1: visual angle amendment is carried out to sequence of video images;
Step S2: illumination amendment is carried out;
Step S3: potential target region candidate frame extracts;
Step S4: scale amendment is carried out;
Step S5: event differentiation is carried out by Harr-like statistical learning classifier.
2. improved complex environment vehicle characteristics according to claim 1 extract and parking method of discrimination, it is characterised in that:
The step S1 includes following sub-step:
Step S11: region of interesting extraction is carried out;
Step S12: zone viewing angle differentiation is carried out;
Step S13: Image Reversal is carried out.
3. improved complex environment vehicle characteristics according to claim 1 extract and parking method of discrimination, special
Sign is: the step 2 includes following sub-step:
Step S21: single Gaussian Background modeling;
Step S22:ROI regional luminance differentiates;
Step S23: Gamma transformation is carried out.
4. improved complex environment vehicle characteristics according to claim 1 extract and parking method of discrimination, it is characterised in that:
The step S3 includes following sub-step:
Step S31: using the mixed Gaussian background modeling of low context update rate;
Step S32: object discrimination is stood;
Step S33: prospect two-value picture dilation erosion;
Step S34: it is excluded based on the local hot spot that Scanny is extracted;
Step S35: output potential target candidate region.
5. improved complex environment vehicle characteristics according to claim 1 extract and parking method of discrimination, it is characterised in that:
The step S4 includes following sub-step:
Step S41: load Harr-Like feature Weak Classifier;
Step S42: potential target region size-varied window traversal;
Step S43: detection window feature remaps.
6. improved complex environment vehicle characteristics according to claim 1 extract and parking method of discrimination, it is characterised in that:
The step S5 includes following sub-step:
Step S51: each Weak Classifier Harr-like characteristic value size is exported;
Step S52: Weak Classifier Nearest Neighbor with Weighted Voting is used;
Step S53: output test result.
7. improved complex environment vehicle characteristics according to claim 1 extract and parking method of discrimination, it is characterised in that:
The method is carried out using the cooperation of improved Adaboost feature learning system, comprising:
Early period rejects atypical characteristics, in the training process of each Weak Classifier, traverses to all Harr-like features
Whether before screening, prejudging first is atypical characteristics, if it is, not carrying out operation to the feature of the type;
For partial points bulk feature existing for vehicle under highway scene, the center contour feature of mutative scale is designed, is allowed to
It participates in during Adaboost feature learning, the optimal option of classification capacity in the center contour feature is obtained by algorithm, is mentioned
High-class effect.
8. improved complex environment vehicle characteristics extract and parking judgement system, it is characterised in that: the system comprises:
Visual angle correction module: the image of different scenes is subjected to visual angle unification, it is special to reduce Harr-like caused by the difference of visual angle
Levy difference in class;
Illumination correction module: by detecting real-time luminosity, brightness regulation is carried out to image, reduces Harr- caused by illumination variation
Difference in like feature class;
Potential target region candidate frame extraction module: foreground target candidate region is got by background subtraction and target scale is believed
Breath reduces unnecessary calculation resources waste when sliding detection window traversal, improves detection real-time, while being next step scale
Amendment provides candidate region information;
Scale correction module: mutative scale detection window traversal, the good trained classifier text of load are carried out to candidate region
Part carries out Harr-like feature in the feature replay of detection window to the Harr-like feature of wherein each Weak Classifier
It penetrates, corresponding Harr-like characteristic value is calculated;
Harr-like characteristic statistics learn discrimination module: special to multiple distinctive Harr-like of Weak Classifier obtained in the previous step
Value indicative is differentiated, obtains final judging result by Nearest Neighbor with Weighted Voting.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910464427.XA CN110188693B (en) | 2019-05-30 | 2019-05-30 | Improved complex environment vehicle feature extraction and parking discrimination method |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910464427.XA CN110188693B (en) | 2019-05-30 | 2019-05-30 | Improved complex environment vehicle feature extraction and parking discrimination method |
Publications (2)
Publication Number | Publication Date |
---|---|
CN110188693A true CN110188693A (en) | 2019-08-30 |
CN110188693B CN110188693B (en) | 2023-04-07 |
Family
ID=67719068
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910464427.XA Active CN110188693B (en) | 2019-05-30 | 2019-05-30 | Improved complex environment vehicle feature extraction and parking discrimination method |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110188693B (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112070172A (en) * | 2020-09-11 | 2020-12-11 | 湖南人文科技学院 | Abnormal target detection method and device based on target analysis and computer equipment |
CN113415232A (en) * | 2021-07-07 | 2021-09-21 | 深圳前海汉视科技有限公司 | Vehicle lamp system for obstacle identification |
Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102637257A (en) * | 2012-03-22 | 2012-08-15 | 北京尚易德科技有限公司 | Video-based detection and recognition system and method of vehicles |
US20150055821A1 (en) * | 2013-08-22 | 2015-02-26 | Amazon Technologies, Inc. | Multi-tracker object tracking |
WO2015147764A1 (en) * | 2014-03-28 | 2015-10-01 | Kisa Mustafa | A method for vehicle recognition, measurement of relative speed and distance with a single camera |
CN104992160A (en) * | 2015-07-16 | 2015-10-21 | 山东大学 | Night preceding vehicle detection method for heavy-duty truck |
CN105373794A (en) * | 2015-12-14 | 2016-03-02 | 河北工业大学 | Vehicle license plate recognition method |
CN105608431A (en) * | 2015-12-22 | 2016-05-25 | 杭州中威电子股份有限公司 | Vehicle number and traffic flow speed based highway congestion detection method |
CN107316036A (en) * | 2017-06-09 | 2017-11-03 | 广州大学 | A kind of insect recognition methods based on cascade classifier |
CN109299672A (en) * | 2018-09-05 | 2019-02-01 | 重庆大学 | The Parking detection system and method for automatic adjusument threshold value and algorithm structure |
-
2019
- 2019-05-30 CN CN201910464427.XA patent/CN110188693B/en active Active
Patent Citations (8)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN102637257A (en) * | 2012-03-22 | 2012-08-15 | 北京尚易德科技有限公司 | Video-based detection and recognition system and method of vehicles |
US20150055821A1 (en) * | 2013-08-22 | 2015-02-26 | Amazon Technologies, Inc. | Multi-tracker object tracking |
WO2015147764A1 (en) * | 2014-03-28 | 2015-10-01 | Kisa Mustafa | A method for vehicle recognition, measurement of relative speed and distance with a single camera |
CN104992160A (en) * | 2015-07-16 | 2015-10-21 | 山东大学 | Night preceding vehicle detection method for heavy-duty truck |
CN105373794A (en) * | 2015-12-14 | 2016-03-02 | 河北工业大学 | Vehicle license plate recognition method |
CN105608431A (en) * | 2015-12-22 | 2016-05-25 | 杭州中威电子股份有限公司 | Vehicle number and traffic flow speed based highway congestion detection method |
CN107316036A (en) * | 2017-06-09 | 2017-11-03 | 广州大学 | A kind of insect recognition methods based on cascade classifier |
CN109299672A (en) * | 2018-09-05 | 2019-02-01 | 重庆大学 | The Parking detection system and method for automatic adjusument threshold value and algorithm structure |
Non-Patent Citations (3)
Title |
---|
叶锋等: "基于Adaboost与背景差分级联的室内人数统计方法", 《福建师范大学学报(自然科学版)》 * |
崔华等: "多特征多阈值级联AdaBoost行人检测器", 《交通运输工程学报》 * |
胡斌等: "计算机视觉在室外移动机器人中的应用", 《自动化学报》 * |
Cited By (4)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112070172A (en) * | 2020-09-11 | 2020-12-11 | 湖南人文科技学院 | Abnormal target detection method and device based on target analysis and computer equipment |
CN112070172B (en) * | 2020-09-11 | 2023-12-22 | 湖南人文科技学院 | Abnormal target detection method and device based on targeting analysis and computer equipment |
CN113415232A (en) * | 2021-07-07 | 2021-09-21 | 深圳前海汉视科技有限公司 | Vehicle lamp system for obstacle identification |
CN113415232B (en) * | 2021-07-07 | 2022-02-01 | 深圳前海汉视科技有限公司 | Vehicle lamp system for obstacle identification |
Also Published As
Publication number | Publication date |
---|---|
CN110188693B (en) | 2023-04-07 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US20230289979A1 (en) | A method for video moving object detection based on relative statistical characteristics of image pixels | |
Huang et al. | A new building extraction postprocessing framework for high-spatial-resolution remote-sensing imagery | |
CN107833221A (en) | A kind of water leakage monitoring method based on multi-channel feature fusion and machine learning | |
CN103035013B (en) | A kind of precise motion shadow detection method based on multi-feature fusion | |
CN109255350B (en) | New energy license plate detection method based on video monitoring | |
CN106373146B (en) | A kind of method for tracking target based on fuzzy learning | |
CN105260749B (en) | Real-time target detection method based on direction gradient binary pattern and soft cascade SVM | |
CN104978567B (en) | Vehicle checking method based on scene classification | |
CN110276284A (en) | Flame identification method, device, equipment and storage medium based on video quality assessment | |
CN101393603A (en) | Method for recognizing and detecting tunnel fire disaster flame | |
CN101094413A (en) | Real time movement detection method in use for video monitoring | |
CN111325769A (en) | Target object detection method and device | |
CN109255326B (en) | Traffic scene smoke intelligent detection method based on multi-dimensional information feature fusion | |
CN113592911B (en) | Apparent enhanced depth target tracking method | |
CN110347870A (en) | The video frequency abstract generation method of view-based access control model conspicuousness detection and hierarchical clustering method | |
CN109063619A (en) | A kind of traffic lights detection method and system based on adaptive background suppression filter and combinations of directions histogram of gradients | |
JP5103665B2 (en) | Object tracking device and object tracking method | |
CN111914634A (en) | Complex-scene-interference-resistant automatic manhole cover type detection method and system | |
CN108256462A (en) | A kind of demographic method in market monitor video | |
CN108921857A (en) | A kind of video image focus area dividing method towards monitoring scene | |
CN112001299B (en) | Tunnel vehicle finger device and lighting lamp fault identification method | |
CN108921215A (en) | A kind of Smoke Detection based on local extremum Symbiotic Model and energy spectrometer | |
CN109190455A (en) | Black smoke vehicle recognition methods based on Gaussian Mixture and autoregressive moving-average model | |
CN110188693A (en) | Improved complex environment vehicle characteristics extract and parking method of discrimination | |
CN110427815A (en) | Realize the method for processing video frequency and device of the effective contents interception of gate inhibition |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |