CN109886086A - Pedestrian detection method based on HOG feature and Linear SVM cascade classifier - Google Patents
Pedestrian detection method based on HOG feature and Linear SVM cascade classifier Download PDFInfo
- Publication number
- CN109886086A CN109886086A CN201910006652.9A CN201910006652A CN109886086A CN 109886086 A CN109886086 A CN 109886086A CN 201910006652 A CN201910006652 A CN 201910006652A CN 109886086 A CN109886086 A CN 109886086A
- Authority
- CN
- China
- Prior art keywords
- pedestrian
- sample
- classifier
- svm
- picture
- 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
Abstract
The invention discloses a kind of pedestrian detection method based on HOG feature and Linear SVM cascade classifier is mainly used for implementing the accuracy of pedestrian detection in the process of moving by improving intelligent vehicle, to solve the safety problem of intelligent vehicle in the process of moving.In order to break it is traditional based on the pedestrian detection algorithm of HOG and SVM when extracting pedestrian HOG feature, pedestrian must generally be kept upright the limitation of posture, the present invention proposes pedestrian being divided into standing, squat down, three kinds of different limb actions of bending over, corresponding Weak Classifier is trained to respectively as a kind of positive sample data set, three obtained Weak Classifier is integrated into a strong classifier again, and using the strong classifier of the cascade structure as the method for intelligent vehicle pedestrian detection model.
Description
Technical field
The present invention relates to a kind of to utilize the cascade classifier based on HOG feature Multi-classifers integrated under intelligent driving environment
The method for carrying out pedestrian detection, is mainly used for solving the safety problem of road driving under unmanned environment, belongs to intelligent driving
Technology and computer vision technique.
Background technique
In recent years, although continuing to bring out out a large amount of algorithm and improved technology in pedestrian's context of detection, and INRIA,
It is tested on the public datas collection such as Caltech, the detection accuracy of pedestrian detection steps up, but due to pedestrian detection itself
Existing some difficulties, for example, pedestrian limb action variation and pedestrian block situation shadow caused by pedestrian detection of equal complexity
It rings, so pedestrian detection still has very big challenge, while the above problem is also urgently to be resolved.At present in pedestrian's test problems
On method there are mainly two types of: first is that be based on traditional algorithm, need hand-designed feature, design process is more complicated；Second is that
Based on deep learning algorithm, needed comprising a series of activities such as the network optimization, parameter adjustment perfect.
Summary of the invention
Goal of the invention: in order to overcome the deficiencies in the prior art, the present invention is provided under a kind of intelligent driving environment
Pedestrian detection method based on HOG feature and Linear SVM cascade classifier is used for safe and accurate, efficient pedestrian detection and row
People's label.The pedestrian's data for passing through intelligent car body sensor collection running environment first, are instructed further according to based on HOG and Linear SVM
The cascade classifier perfected carries out the processing of data to detect the pedestrian on road, and prompts vehicle driver or pass through intelligence
It can control and achieve the purpose that avoid pedestrian.This method stands according to pedestrian, squats down, bends over the three of three kinds of limb action characteristic Designs
Kind classifier is simultaneously integrated, and three kinds of limb action features are respectively as positive sample, in one of limb action feature of training
As positive sample data set when, it is three points available in this way using other two kinds of positive samples and background image as negative sample
Class device finally by these three classifiers combinations at the classifier of a cascade structure, and is written in intelligent driving system.Intelligence
Control loop marks pedestrian target after can detecting identification pedestrian by above-mentioned cascade classifier again, ensure that intelligent vehicle is expert at
The accuracy of pedestrian detection is carried out during sailing.
Technical solution: to achieve the above object, the technical solution adopted by the present invention are as follows:
The present invention is directed to improve the accuracy of intelligent vehicle pedestrian detection in the process of moving, key is intelligent driving system
Real-time, high efficiency, the accuracy of pedestrian detection algorithm in system.Firstly the need of the sample for collecting pedestrian on a large amount of travel
Collection, is divided into positive sample collection and negative sample collection for sample set.If positive sample collection is the pedestrian to stand, the pedestrian that squats down bends over
Pedestrian and background are negative sample set；If positive sample collection is the pedestrian to squat down, the pedestrian to stand, the pedestrian to bend over and background are negative
Sample set；If positive sample collection is the pedestrian to bend over, the pedestrian of the pedestrian, standing that squat down and background are negative sample set.It counts respectively
The HOG feature of positive sample collection and negative sample collection is calculated, and is trained respectively with Linear SVM, so that three classifiers are obtained, it will
These three classifiers combinations are again trained at the classifier of a cascade structure and with Adaboost algorithm setting weight.Its
In, the weight of three classifiers, certain limb of pedestrian can be set according to pedestrian's probability that limb action occurs in the process of walking
The probability that body movement occurs is bigger, and weight setting is higher.Such as standing activities are that normal limbs are dynamic in the process of walking by pedestrian
Make, thus with the pedestrian of standing be positive sample set classifier setting weight can be higher.Then the cascade point training completed
Class device is written in intelligent driving system, is input with pedestrian's data in intelligent vehicle ambient enviroment, with the intelligence of intelligent vehicle
Prompt or intelligent control are output, to achieve the purpose that avoid pedestrian.By marking row again after detecting identification pedestrian layer by layer
The security strategy of people's target can substantially reduce the false detection rate of pedestrian detection, improve the safety of intelligent vehicle in the process of moving
Property.
A kind of pedestrian detection method based on HOG feature and Linear SVM cascade classifier, includes the following steps:
Step 1: pedestrian's picture formation sample set that acquisition stands, squats down and bend over three kinds of limb actions, by sample set point
For three classes:
Sample data set T1: using the pedestrian's picture for limb action of standing as positive sample, squat down and limb action of bending over
Pedestrian's picture is as negative sample；
Sample data set T2: using the pedestrian's picture for limb action of squatting down as positive sample, stand and limb action of bending over
Pedestrian's picture is as negative sample；
Sample data set T3: using the pedestrian's picture for limb action of bending over as positive sample, limb action of standing and squat down
Pedestrian's picture is as negative sample；
In sample data set T1、T2、T3In, label α is assigned to each positive sample, assigns label β to each negative sample；
Step 2: the HOG feature of each sample is extracted；
Step 3: setting initial weight, by sample data set T1、T2、T3In positive negative sample first trained with initial weight
One SVM Weak Classifier S1；Then weight is updated according to learning error rate, by sample data set T1、T2、T3In positive negative sample
Go out a SVM Weak Classifier S with weight re -training is updated2；Then weight is updated according to new learning error rate, by sample number
According to collection T1、T2、T3In positive negative sample update weight re -training go out a SVM Weak Classifier S3；It so repeats, until SVM
The quantity of Weak Classifier, which reaches, to be specified number；
Step 4: the SVM Weak Classifier specified number is carried out by weight training based on Adaboost thought, and assembles one
A cascade of strong classifiers；
Step 5: using cascade of strong classifiers only to sample data set T1、T2、T3In negative sample identified, by mistake
The image of identification is added to negative sample difficulty example concentration；Training is optimized to cascade of strong classifiers using negative sample difficulty example collection.
Specifically, other than acquiring pedestrian's picture, while acquiring background picture in the step 1, background picture made
For sample data set T1、T2、T3Middle negative sample.
Specifically, in the step 2, the HOG feature extracting method of sample are as follows: gray proces are carried out to picture first；So
Gamma correction is carried out to the picture after gray proces afterwards；HOG feature extraction finally is carried out to the picture after Gamma correction, first will
If picture is divided into stem cell units, then acquires the direction histogram at the gradient of each pixel or edge in cell factory, then will be square
Form the HOG feature of sample altogether to set of histograms.
Specifically, in the step 3, the training method of SVM Weak Classifier are as follows: by sample data set T1、T2、T3In just
The HOG feature and label of negative sample, which are input in SVM, to be trained, and finally trained result will save as XML file, XML
The content of file includes two arrays and a floating number, and two number groups are denoted as Support Vector and Alpha, floating-point respectively
Number scale is Rho.
Specifically, in the step 4, cascade of strong classifiers are as follows:
Wherein: SkIt (x) is k-th of SVM Weak Classifier, αkFor Sk(x) weight, K are the sum of SVM Weak Classifier, sign
() is sign function, and x is data set；ekFor error rate.
Using intelligent automobile of the invention, pedestrian detection method under intelligent driving environment the following steps are included:
(1) intelligent vehicle in the process of moving, perceives peripheral ring by camera, radar or multi-sensor fusion technology
Border provides surrounding enviroment data and is written in intelligent driving system；
(2) data that intelligent driving system is imported according to sensor detect identification using cascade classifier described above
Mark pedestrian target after pedestrian again, and shown with patterned interface to driver, or directly pass through intelligent driving system to
The control instruction of intelligent control layer transmitting manipulation intelligent vehicle；
(3) pedestrian's specific location that last intelligent vehicle is shown according to intelligent driving system, provides the necessary safety of driver
Property prompt, such as: whistle is illustrated, sounds an alarm, or is subject to intelligent control to avoid the pedestrian detected, such as: lane change,
Acceleration-deceleration etc..
The utility model has the advantages that the pedestrian detection method provided by the invention based on HOG feature and Linear SVM cascade classifier, phase
For the prior art, have the advantage that 1, for the HOG feature for extracting pedestrian when, pedestrian must generally be kept upright appearance
The limitation of gesture, pedestrian's posture is divided into standing, squatted down, three kinds of limb action features of bending over by the present invention, and the line designed according to this
The method of property SVM cascade classifier can substantially reduce the omission factor of pedestrian detection；2, the missing inspection of pedestrian is that intelligent vehicle is travelling
Occur one of important hidden danger of safety problem in the process, by improving the pedestrian detection algorithm in intelligent driving system, improves row
The accuracy rate of people's detection, is able to ascend the safety of intelligent vehicle in the process of moving；3, the pedestrian detection based on deep learning is calculated
Method is although high-efficient, but the algorithm is all much higher than traditional algorithm to trained data volume and on-board processing machine performance requirement,
Therefore the improvement of traditional algorithm is beneficial to save the manufacturing cost of intelligent vehicle.
Detailed description of the invention
Fig. 1 is the realization procedure chart of HOG feature extraction algorithm；
Fig. 2 is that Linear SVM solves flow chart；
Fig. 3 is the basic ideas flow chart of Adaboost algorithm；
Fig. 4 is pedestrian detection flow chart in intelligent driving.
Specific embodiment
The present invention will be further explained with reference to the accompanying drawing.
Pedestrian detection technology in intelligent driving technology is intelligent driving system for given image and video, judgement
Wherein whether there is pedestrian out, and provides the technology of the specific location of pedestrian.Its key is pedestrian detection algorithm, in the present invention
In, pedestrian detection algorithm, which mainly passes through HOG feature extraction, SVM training and cascade classifier three steps of training, to be completed.Compared to being based on
The pedestrian detection method of deep learning, the present invention use the pedestrian detection side based on HOG feature and Linear SVM cascade classifier
Method can effectively solve the problems, such as to cause discrimination to reduce because pedestrian's limb action is excessive, and advantage is even more to be the letter of its model
Single, training is convenient, flexible.
Wherein histograms of oriented gradients (Histogram of Oriented Gradient, abbreviation HOG) is characterized in one kind
It is used to carry out the feature description value of object detection in computer vision and image procossing, essence is the statistical information of gradient.
Its main thought is in a sub-picture, because gradient is primarily present in the place at edge, the presentation and shape of localized target
Shape can be described well by gradient, therefore we can constitute target by calculating with the gradient of statistical picture regional area
Feature.Compared with other character description methods, HOG has the advantage of itself.For example, we are very on big space field
Be hard to keep the geometric deformation of image and the invariance of optical deformation, but HOG is operated on the local pane location of image, institute
Good invariance can be kept to the deformation of image geometry and optical deformation with it.But the extraction of HOG feature is only applicable in
Under the conditions of rough airspace sampling, fine direction sampling and the normalization of stronger indicative of local optical etc., therefore retouched with HOG
When stating the feature of pedestrian, the posture that pedestrian must generally be kept upright if the limb action of pedestrian is excessive, such as squats down, is curved
Waist etc. will seriously affect the effect of feature description.
Support vector machines (Support Vector Machine, abbreviation SVM) is an optimal classification algorithm.It is so-called optimal
Classification exactly requires classification line to be not only able to two classes are faultless separated, and the class interval between two classes is maximum, promotes
To higher dimensional space, optimal classification line just becomes optimal classification surface.SVM is trained using the thought of class interval, it
Dependent on the pretreatment to data, so firstly the need of the space expression raw mode in more higher-dimension, then in this higher-dimension sky
A Nonlinear Mapping is found between as soon as, the initial data for being belonging respectively to two classes in this way can be separated by a hyperplane.It will count
It is exactly to find an optimal classifying face so that class interval is maximum according to the purpose for being mapped to higher dimensional space from luv space.
In addition algorithm of support vector machine can use different kernel functions, the main linear kernel function of kernel function (Linear) and Gauss
Kernel function (RBF) utilizes use using the classifier for using the SVM algorithm of linear kernel function to be trained for linear classifier
The classifier that the SVM algorithm of gaussian kernel function is trained is Nonlinear Classifier.The anticipation function of linear classifier is simple, point
Class speed is fast, and supporting vector number can be very more in higher dimensional space for Nonlinear Classifier, and classification speed is far below linear classification
Device.Because pedestrian's intrinsic dimensionality that HOG is extracted is higher, selecting linear classifier is a preferable selection.
Cascade (Cascade) structure classifier be a kind of efficient classifier, be by several simple Weak Classifiers by
It is combined into a hierarchical classifier according to certain order, to substitute the monolithic devices classifier an of bulky complex.It is cascading
Sub-fraction key feature is selected from huge feature set with Adaboost algorithm in the structure of classifier, is weighed by setting
Retraining Weak Classifier is combined into strong classifier further according to the weight of Weak Classifier, is finally reached the classification effect for improving verification and measurement ratio
Fruit.
The training of cascade classifier based on HOG feature and Linear SVM calculates the HOG description of positive and negative sample image first
Value, forms an eigenvectors matrix, each feature vector has a category vector, and category vector may indicate that feature vector
Classification；Then it is trained using the method for Linear SVM；Trained three Linear SVM classifiers save as XML file, and
Weight training is carried out to three classifiers with Adaboost algorithm, formation is well-bedded, can be used to carry out pedestrian detection
Cascade classifier.It is finally trained again in conjunction with difficult example, so-called hardly possible example refers to using the classifier of training for the first time in negative sample
All rectangle frames detected when pedestrian detection are carried out in original image, these rectangle frame regions are all wrong reports, by the rectangle frame of wrong report
Picture is saved as, is added in initial negative sample set, is re-started the training of Linear SVM, integrate again.
One, feature extraction
Sample data set main source INRIA.Positive sample collection is obtained, and extracts positive sample and the HOG feature of pedestrian is concentrated to retouch
State value.Positive sample image uses the standing of 96*160 pixel, the pedestrian's picture squatted down or bent over.In use, picture is left up and down
16 pixels are all removed on the right side, intercept the human body image part of intermediate 64*128 pixel.Negative sample collection is obtained, and extracts negative sample
The HOG feature description value of collection, the available image without detection target of negative sample image are cut out to obtain at random, and size is equally 64*
128 pixels.Usual negative sample quantity will be far longer than positive sample number.Positive and negative sample set is divided into three classes:
Sample data set T1: using the pedestrian's picture for limb action of standing as positive sample, squat down and limb action of bending over
Pedestrian's picture and background picture are as negative sample；
Sample data set T2: using the pedestrian's picture for limb action of squatting down as positive sample, stand and limb action of bending over
Pedestrian's picture and background picture are as negative sample；
Sample data set T3: using the pedestrian's picture for limb action of bending over as positive sample, limb action of standing and squat down
Pedestrian's picture and background picture are as negative sample；
In sample data set T1、T2、T3In, label 1 is assigned to each positive sample, assigns label 0 to each negative sample；
The HOG feature extraction of image is divided into two steps, as shown in Figure 1:
Step 1: image preprocessing is to do color space normalized to the image being collected into, reaches noise reduction effect；
Include two steps:
(1) RGB component is converted to gray level image that is, for color image by image gray processing, and the method used is weighting
Mean value method.Different weights is assigned to three components of image first, is then weighted and averaged, enabling the value is gray value,
Expression formula are as follows:
F (i, j)=VR.R(i,j)+VG.G(i.j)+VB.B(i,j)
According to the available final gray processing formula of the weight being arranged as follows are as follows:
Gray=0.3 × R+0.59 × G+0.11 × B
(2) Gamma correct, in the uneven situation of brightness of image, corrected by Gamma, can by overall brightness be turned up or
Person reduces.The method that Gamma correction uses is square-root method, the formula of square root method are as follows:
Y (x, y)=I (x, y)γ
Step 2: HOG feature extraction
Small connected region is divided the image into first, is cell factory it.Then each pixel in cell factory is acquired
Gradient or edge direction histogram.These set of histograms finally can be formed by feature description value altogether.Specific step
It is rapid as follows:
(1) gradient of image abscissa and ordinate direction is calculated, and calculates the gradient direction of each location of pixels accordingly
Value, derivation operations can not only capture profile, the shadow and some texture informations, moreover it is possible to the influence that further weakened light shines.
(2) coding method is provided for local image region, keep to the posture of human object in image and appearance
Hyposensitiveness perception.
(3) variation for the variation and foreground-background contrast shone due to local light, so that the variation range of gradient intensity
Very big, this just needs to normalize gradient intensity.Normalization can further press illumination, shade and edge
Contracting.
(4) final step is exactly that will test the collection of the part progress HOG feature of all overlappings in window, and they are tied
Final feature vector is synthesized to use for training classifier.
Two, SVM training
By sample data set T1、T2、T3In positive negative sample a SVM Weak Classifier S is first trained with initial weight1。
The training method of SVM Weak Classifier are as follows: by sample data set T1、T2、T3In positive negative sample HOG feature and label input
It is trained into SVM, the principle flow chart of SVM is as shown in Figure 2；Finally trained result will save as XML file, XML text
The content of part includes two arrays and a floating number.
Then weight is updated according to learning error rate, so that SVM Weak Classifier S1The high training sample of learning error rate
Weight is got higher, and the sample that these error rates can be made high so is more paid attention in the Weak Classifier below；By sample
Notebook data collection T1、T2、T3In positive negative sample update weight re -training go out a SVM Weak Classifier S2；Then according to new
Learning error rate updates weight, by sample data set T1、T2、T3In positive negative sample update weight re -training go out a SVM
Weak Classifier S3。
Three, cascade classifier training
Three Weak Classifiers of above-mentioned steps generation, basic ideas flow chart such as Fig. 3 are integrated with Adaboost algorithm
It is shown.This 3 Weak Classifiers are integrated by aggregation policy finally, obtain final strong learner.Adaboost set
Strategy is using weighted mean method, final strong classifier are as follows:
Wherein: SkIt (x) is k-th of SVM Weak Classifier, αkFor Sk(x) weight, K are the sum of SVM Weak Classifier, sign
() is sign function, and x is sample set；ekFor error rate.
Four, in conjunction with difficult example re -training
Optimize the classifier of cascade structure described above with Boost algorithm, the basic thought of boot strap is to make first
A model is trained with original negative sample set, then collects the negative sample classified by this initial model mistake, forms one
Negative sample difficulty example collection, with the new model of this negative sample difficulty example collection training.
The cascade classifier of optimization can be applied in the pedestrian detection module in intelligent driving, the row comprising cascade classifier
The workflow of people's detection module is as shown in Figure 4.
The above is only a preferred embodiment of the present invention, it should be pointed out that: for the ordinary skill people of the art
For member, various improvements and modifications may be made without departing from the principle of the present invention, these improvements and modifications are also answered
It is considered as protection scope of the present invention.
Claims (5)
1. a kind of pedestrian detection method based on HOG feature and Linear SVM cascade classifier, it is characterised in that: including walking as follows
It is rapid:
Step 1: sample set is divided into three by pedestrian's picture formation sample set that acquisition stands, squats down and bend over three kinds of limb actions
Class:
Sample data set T1: using the pedestrian's picture for limb action of standing as positive sample, squats down and the pedestrian for limb action of bending over schemes
Piece is as negative sample；
Sample data set T2: using the pedestrian's picture for limb action of squatting down as positive sample, stands and the pedestrian for limb action of bending over schemes
Piece is as negative sample；
Sample data set T3: using the pedestrian's picture for limb action of bending over as positive sample, the pedestrian for limb action of standing and squat down schemes
Piece is as negative sample；
In sample data set T1、T2、T3In, label α is assigned to each positive sample, assigns label β to each negative sample；
Step 2: the HOG feature of each sample is extracted；
Step 3: setting initial weight, by sample data set T1、T2、T3In positive negative sample one is first trained with initial weight
SVM Weak Classifier S1；Then weight is updated according to learning error rate, by sample data set T1、T2、T3In positive negative sample with more
New weight re -training goes out a SVM Weak Classifier S2；Then weight is updated according to new learning error rate, by sample data set
T1、T2、T3In positive negative sample update weight re -training go out a SVM Weak Classifier S3；It so repeats, until weak point of SVM
The quantity of class device, which reaches, to be specified number；
Step 4: the SVM Weak Classifier specified number is carried out by weight training based on Adaboost thought, and assembles a grade
Join strong classifier；
Step 5: using cascade of strong classifiers only to sample data set T1、T2、T3In negative sample identified, by wrong identification
Image be added to negative sample difficulty example concentration；Training is optimized to cascade of strong classifiers using negative sample difficulty example collection.
2. the pedestrian detection method according to claim 1 based on HOG feature and Linear SVM cascade classifier, feature
It is: in the step 1, other than acquiring pedestrian's picture, while background picture is acquired, using background picture as sample data
Collect T1、T2、T3Middle negative sample.
3. the pedestrian detection method according to claim 1 based on HOG feature and Linear SVM cascade classifier, feature
It is: in the step 2, the HOG feature extracting method of sample are as follows: gray proces are carried out to picture first；Then to gray scale at
Picture after reason carries out Gamma correction；HOG feature extraction finally is carried out to the picture after Gamma correction, if being first divided into picture
Stem cell units, then acquire the direction histogram at the gradient of each pixel or edge in cell factory, then by direction histogram group
The HOG feature of sample is formed altogether.
4. the pedestrian detection method according to claim 1 based on HOG feature and Linear SVM cascade classifier, feature
It is: in the step 3, the training method of SVM Weak Classifier are as follows: by sample data set T1、T2、T3In positive negative sample
HOG feature and label, which are input in SVM, to be trained, and finally trained result will save as XML file, XML file it is interior
Hold includes two arrays and a floating number.
5. the pedestrian detection method according to claim 1 based on HOG feature and Linear SVM cascade classifier, feature
It is: in the step 4, cascade of strong classifiers are as follows:
Wherein: SkIt (x) is k-th of SVM Weak Classifier, αkFor Sk(x) weight, K are the sum of SVM Weak Classifier, sign ()
For sign function, x is data set；ekFor error rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910006652.9A CN109886086B (en) | 2019-01-04 | 2019-01-04 | Pedestrian detection method based on HOG (histogram of oriented gradient) features and linear SVM (support vector machine) cascade classifier |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201910006652.9A CN109886086B (en) | 2019-01-04 | 2019-01-04 | Pedestrian detection method based on HOG (histogram of oriented gradient) features and linear SVM (support vector machine) cascade classifier |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN109886086A true CN109886086A (en) | 2019-06-14 |
| CN109886086B CN109886086B (en) | 2020-12-04 |
Family
ID=66925514
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201910006652.9A Active CN109886086B (en) | 2019-01-04 | 2019-01-04 | Pedestrian detection method based on HOG (histogram of oriented gradient) features and linear SVM (support vector machine) cascade classifier |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN109886086B (en) |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516523A (en) * | 2019-06-18 | 2019-11-29 | 中国矿业大学 | Danger zone pedestrian detection method in mobile video inspection |
| CN111060904A (en) * | 2019-12-25 | 2020-04-24 | 中国汽车技术研究中心有限公司 | Blind area monitoring method based on millimeter wave and vision fusion perception |
| CN111080524A (en) * | 2019-12-19 | 2020-04-28 | 吉林农业大学 | Plant disease and insect pest identification method based on deep learning |
| CN111783604A (en) * | 2020-06-24 | 2020-10-16 | 中国第一汽车股份有限公司 | Vehicle control method, device and equipment based on target identification and vehicle |
| CN114463813A (en) * | 2022-01-21 | 2022-05-10 | 苏州浪潮智能科技有限公司 | Expression recognition method, system and related device based on HOG characteristics |
Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008039514A1 (en) * | 2006-09-26 | 2008-04-03 | Cytrx Corporation | Pharmaceutical compositions and methods for treating diseases associated with neurodegeneration |
| CN102663409A (en) * | 2012-02-28 | 2012-09-12 | 西安电子科技大学 | Pedestrian tracking method based on HOG-LBP |
| CN103211599A (en) * | 2013-05-13 | 2013-07-24 | 桂林电子科技大学 | Method and device for monitoring tumble |
| CN104318263A (en) * | 2014-09-24 | 2015-01-28 | 南京邮电大学 | Real-time high-precision people stream counting method |
| CN104463232A (en) * | 2014-12-30 | 2015-03-25 | 中山大学 | Density crowd counting method based on HOG characteristic and color histogram characteristic |
| CN104484680A (en) * | 2014-09-26 | 2015-04-01 | 徐晓晖 | Multi-model multi-threshold combined pedestrian detection method |
| CN104715238A (en) * | 2015-03-11 | 2015-06-17 | 南京邮电大学 | Pedestrian detection method based on multi-feature fusion |
| CN106339684A (en) * | 2016-08-26 | 2017-01-18 | 北汽福田汽车股份有限公司 | Pedestrian detection method, device and vehicle |
| CN106934380A (en) * | 2017-03-19 | 2017-07-07 | 北京工业大学 | A kind of indoor pedestrian detection and tracking based on HOG and MeanShift algorithms |
| CN106971158A (en) * | 2017-03-23 | 2017-07-21 | 南京邮电大学 | A kind of pedestrian detection method based on CoLBP symbiosis feature Yu GSS features |
| US20170213081A1 (en) * | 2015-11-19 | 2017-07-27 | Intelli-Vision | Methods and systems for automatically and accurately detecting human bodies in videos and/or images |
| CN107122726A (en) * | 2017-04-19 | 2017-09-01 | 高新兴科技集团股份有限公司 | A kind of multi-pose pedestrian detection method |
-
2019
- 2019-01-04 CN CN201910006652.9A patent/CN109886086B/en active Active
Patent Citations (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2008039514A1 (en) * | 2006-09-26 | 2008-04-03 | Cytrx Corporation | Pharmaceutical compositions and methods for treating diseases associated with neurodegeneration |
| CN102663409A (en) * | 2012-02-28 | 2012-09-12 | 西安电子科技大学 | Pedestrian tracking method based on HOG-LBP |
| CN103211599A (en) * | 2013-05-13 | 2013-07-24 | 桂林电子科技大学 | Method and device for monitoring tumble |
| CN104318263A (en) * | 2014-09-24 | 2015-01-28 | 南京邮电大学 | Real-time high-precision people stream counting method |
| CN104484680A (en) * | 2014-09-26 | 2015-04-01 | 徐晓晖 | Multi-model multi-threshold combined pedestrian detection method |
| CN104463232A (en) * | 2014-12-30 | 2015-03-25 | 中山大学 | Density crowd counting method based on HOG characteristic and color histogram characteristic |
| CN104715238A (en) * | 2015-03-11 | 2015-06-17 | 南京邮电大学 | Pedestrian detection method based on multi-feature fusion |
| US20170213081A1 (en) * | 2015-11-19 | 2017-07-27 | Intelli-Vision | Methods and systems for automatically and accurately detecting human bodies in videos and/or images |
| CN106339684A (en) * | 2016-08-26 | 2017-01-18 | 北汽福田汽车股份有限公司 | Pedestrian detection method, device and vehicle |
| CN106934380A (en) * | 2017-03-19 | 2017-07-07 | 北京工业大学 | A kind of indoor pedestrian detection and tracking based on HOG and MeanShift algorithms |
| CN106971158A (en) * | 2017-03-23 | 2017-07-21 | 南京邮电大学 | A kind of pedestrian detection method based on CoLBP symbiosis feature Yu GSS features |
| CN107122726A (en) * | 2017-04-19 | 2017-09-01 | 高新兴科技集团股份有限公司 | A kind of multi-pose pedestrian detection method |
Cited By (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN110516523A (en) * | 2019-06-18 | 2019-11-29 | 中国矿业大学 | Danger zone pedestrian detection method in mobile video inspection |
| CN110516523B (en) * | 2019-06-18 | 2023-08-25 | 中国矿业大学 | Pedestrian detection method for dangerous area in mobile video inspection |
| CN111080524A (en) * | 2019-12-19 | 2020-04-28 | 吉林农业大学 | Plant disease and insect pest identification method based on deep learning |
| CN111060904A (en) * | 2019-12-25 | 2020-04-24 | 中国汽车技术研究中心有限公司 | Blind area monitoring method based on millimeter wave and vision fusion perception |
| CN111060904B (en) * | 2019-12-25 | 2022-03-15 | 中国汽车技术研究中心有限公司 | Blind area monitoring method based on millimeter wave and vision fusion perception |
| CN111783604A (en) * | 2020-06-24 | 2020-10-16 | 中国第一汽车股份有限公司 | Vehicle control method, device and equipment based on target identification and vehicle |
| CN114463813A (en) * | 2022-01-21 | 2022-05-10 | 苏州浪潮智能科技有限公司 | Expression recognition method, system and related device based on HOG characteristics |
| CN114463813B (en) * | 2022-01-21 | 2024-02-13 | 苏州浪潮智能科技有限公司 | HOG feature-based expression recognition method, system and related device |
Also Published As
| Publication number | Publication date |
|---|---|
| CN109886086B (en) | 2020-12-04 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109886086A (en) | Pedestrian detection method based on HOG feature and Linear SVM cascade classifier | |
| CN104778453B (en) | A kind of night pedestrian detection method based on infrared pedestrian's brightness statistics feature | |
| CN106650731B (en) | Robust license plate and vehicle logo recognition method | |
| CN106980852B (en) | Based on Corner Detection and the medicine identifying system matched and its recognition methods | |
| CN110363140A (en) | A kind of human action real-time identification method based on infrared image | |
| CN107194346A (en) | A kind of fatigue drive of car Forecasting Methodology | |
| CN106682603B (en) | Real-time driver fatigue early warning system based on multi-source information fusion | |
| CN107330453A (en) | The Pornographic image recognizing method of key position detection is recognized and merged based on substep | |
| CN103020992B (en) | A kind of video image conspicuousness detection method based on motion color-associations | |
| CN104598885B (en) | The detection of word label and localization method in street view image | |
| CN107491769A (en) | Method for detecting fatigue driving and system based on AdaBoost algorithms | |
| CN103714659B (en) | Fatigue driving identification system based on double-spectrum fusion | |
| CN109817013A (en) | Parking stall state identification method and device based on video flowing | |
| CN111062292B (en) | Fatigue driving detection device and method | |
| CN109635784A (en) | Traffic sign recognition method based on improved convolutional neural networks | |
| CN103617426A (en) | Pedestrian target detection method under interference by natural environment and shelter | |
| CN106529461A (en) | Vehicle model identifying algorithm based on integral characteristic channel and SVM training device | |
| CN106503748A (en) | A kind of based on S SIFT features and the vehicle targets of SVM training aids | |
| CN111553201A (en) | Traffic light detection method based on YOLOv3 optimization algorithm | |
| CN107092876A (en) | The low-light (level) model recognizing method combined based on Retinex with S SIFT features | |
| CN106203338B (en) | Human eye state method for quickly identifying based on net region segmentation and threshold adaptive | |
| CN109948433A (en) | A kind of embedded human face tracing method and device | |
| CN106485715A (en) | A kind of unstructured road recognition methods | |
| CN107315990A (en) | A kind of pedestrian detection algorithm based on XCS LBP features and cascade AKSVM | |
| CN108573219B (en) | Eyelid key point accurate positioning method based on deep convolutional neural network |
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 | ||
| CB02 | Change of applicant information |
Address after: 210000, 66 new model street, Gulou District, Jiangsu, Nanjing Applicant after: NANJING University OF POSTS AND TELECOMMUNICATIONS Address before: No. 30 Guangyue Road, Qixia Street, Qixia District, Nanjing City, Jiangsu Province, 210000 Applicant before: NANJING University OF POSTS AND TELECOMMUNICATIONS |
|
| CB02 | Change of applicant information | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |