CN110232314A - A kind of image pedestrian's detection method based on improved Hog feature combination neural network - Google Patents
A kind of image pedestrian's detection method based on improved Hog feature combination neural network Download PDFInfo
- Publication number
- CN110232314A CN110232314A CN201910351442.3A CN201910351442A CN110232314A CN 110232314 A CN110232314 A CN 110232314A CN 201910351442 A CN201910351442 A CN 201910351442A CN 110232314 A CN110232314 A CN 110232314A
- Authority
- CN
- China
- Prior art keywords
- image
- neural network
- hog feature
- pedestrian
- region
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000013528 artificial neural network Methods 0.000 title claims abstract description 34
- 238000001514 detection method Methods 0.000 title claims abstract description 28
- 238000000034 method Methods 0.000 claims abstract description 28
- 238000000605 extraction Methods 0.000 claims description 9
- 238000012360 testing method Methods 0.000 claims description 8
- 238000012549 training Methods 0.000 claims description 6
- 230000033001 locomotion Effects 0.000 claims description 5
- 239000000284 extract Substances 0.000 claims description 3
- 238000003062 neural network model Methods 0.000 claims description 3
- 238000007781 pre-processing Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 210000004027 cell Anatomy 0.000 description 10
- 230000006870 function Effects 0.000 description 5
- 230000000694 effects Effects 0.000 description 3
- 230000004913 activation Effects 0.000 description 2
- 238000012544 monitoring process Methods 0.000 description 2
- 210000002569 neuron Anatomy 0.000 description 2
- 229930091051 Arenine Natural products 0.000 description 1
- SWECWXGUJQLXJF-BTJKTKAUSA-N Dimetindene maleate Chemical compound OC(=O)\C=C/C(O)=O.CN(C)CCC=1CC2=CC=CC=C2C=1C(C)C1=CC=CC=N1 SWECWXGUJQLXJF-BTJKTKAUSA-N 0.000 description 1
- 238000013459 approach Methods 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 230000006854 communication Effects 0.000 description 1
- 238000013480 data collection Methods 0.000 description 1
- 238000013499 data model Methods 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 235000013399 edible fruits Nutrition 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000007689 inspection Methods 0.000 description 1
- 230000036544 posture Effects 0.000 description 1
- 230000000644 propagated effect Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2411—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on the proximity to a decision surface, e.g. support vector machines
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V40/00—Recognition of biometric, human-related or animal-related patterns in image or video data
- G06V40/10—Human or animal bodies, e.g. vehicle occupants or pedestrians; Body parts, e.g. hands
- G06V40/103—Static body considered as a whole, e.g. static pedestrian or occupant recognition
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Evolutionary Biology (AREA)
- Evolutionary Computation (AREA)
- Bioinformatics & Computational Biology (AREA)
- General Engineering & Computer Science (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Human Computer Interaction (AREA)
- Multimedia (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of image pedestrian's detection methods based on improved Hog feature combination neural network, and image pixel gradient is calculated by using Steerable filter algorithm, improve arithmetic speed.In addition, when carrying out pedestrian detection being detected with fixed whole image of window sliding, but moving region has first been determined with frame difference method, and moving region is carried out to certain scaling, it is detected in region, the time of operation is reduced, detection efficiency is improved.
Description
Technical field
The present invention relates to the technical fields of Machine Vision Detection, more particularly to one kind to be combined based on improved Hog feature
Image pedestrian's detection method of neural network.
Background technique
Pedestrian detection is to judge to whether there is pedestrian in image or video sequence using computer vision technique and give essence
True positioning.Classical pedestrian detection method uses hog feature that svm classifier is added to realize that this method is taken to image
Using fixed window scanning, each width is scanned and obtains hog feature and the feature using trained svm classifier to extraction
Classify.But since pedestrian wears the different of dress ornament clothes, pedestrian is likely to occur different postures, pedestrian caused by camera position
In the angle of image or video difference, need to collect a large amount of sample.But the svm classifier of (non-linear) with kernel function exists
A large amount of running memory and time can be consumed when handling mass data, and the effect classified is not very good in the case, svm
Classifier is relatively suitble to do small sample problem.Neural network is in a fast-developing status, firstly because in neural network
Feature extraction is incorporated in training process, furthermore, in today of big data era, obtain what hardly possible large data collection is no longer
Topic allows memory and calculated performance " democratization " finally, with advances in technology.Therefore, it is carried out with neural network based on full-page proof
This pedestrian detection is feasible effective.Hog feature is a kind of method extracted Image edge gradient and be, has preferable robust
Property, but in traditional gradient calculates, using relatively simple.
Summary of the invention
It is an object of the invention to overcome the deficiencies of the prior art and provide a kind of detection accuracy height, detection efficiency are high
Image pedestrian's detection method based on Steerable filter Hog feature combination neural network.
To achieve the above object, technical solution provided by the present invention are as follows:
Including following several stages:
The study stage:
S1-1: establishing pedestrian sample library, the non-pedestrian sample of pedestrian sample and half including half;
S1-2: pre-processing the sample in step S1-1, by color image gray processing;
S1-3: to pretreated sample extraction Hog feature vector;
S1-4: establishing neural network, is trained to the Hog feature vector of extraction, obtains network weight parameter;
Test phase:
S2-1: pretreatment identical with the study stage is carried out to picture sample to be detected;
S2-2: the Hog feature vector by the step S2-1 pretreated picture obtained is calculated;
S2-3: the feature vector that step S2-2 is obtained tests the neural network model after training, calculates network
The accuracy of model adjusts network structure by actual result;
Application stage:
S3-1: the region of video moving object is extracted using frame differential method, is zoomed in and out according to the size in region, region
If being less than fixed sliding window, region is amplified to window size, extract the Hog feature neural network in region to determine whether
For pedestrian.
Further, in the step S1-3 and step S2-2, extracting Hog feature vector, specific step is as follows:
A1: use direction controllable filter calculates the gradient of each pixel;
A2: building cell factory lattice simultaneously count its histogram of gradients;
A3: the feature vector of all cell factories of block is together in series and is normalized, obtains the Hog of the block
Feature;
A4: the feature vector of whole blocks is together in series, and obtains the Hog Feature Descriptor of sample.
Further, the step A1 use direction controllable filter calculates the specific steps of the gradient of each pixel such as
Under:
Using the constituted controllable filter of Gauss second dervative, form is as follows:
Wherein,Respectively Gaussian function corresponding direction second dervative,
Expression is as follows:
Coefficient of correspondence is divided into:
The Steerable filter in tectonic level direction and vertical direction, is denoted as F respectively0, Fπ/2, calculating gray level image (x,
Y) the gradient value G of place's pixel both horizontally and verticallyX(x, y), Gy(x, y):
Gx(x, y)=F0*I;
Gy(x, y)=Fπ/2*I;
Wherein, I is gray level image;
Then direction and the amplitude of gradient are calculated separately:
Further, in the step A2, the range of gradient direction is 0~180 degree, is divided into nine parts, i.e. histogram
There are nine sections, each interval range is 20 degree;
Nearest Neighbor with Weighted Voting is taken in histogram ballot, i.e., the gradient magnitude of each pixel obtains one nine dimension as ballot weight
Feature vector.
Further, detailed process is as follows by the step S3-1:
B1: video source is obtained;
B2: moving region is determined using frame differential method;
B3: it is slided using fixed window;
B4: judging whether window has target, if so, then storing target area coordinates;If no, return step B3;
B5: judging whether coverage motion region, if so, terminating;If it is not, then return step B3.
Further, in the step B2, by two frame adjacent in video flowing or it is separated by the two images pixels of a few frame images
Value is subtracted each other, and carries out thresholding to the image after subtracting each other to extract the moving region in image;If subtracting each other the frame number of two field pictures
Respectively kth frame, (k+1) frame, then its frame image is respectively as follows:
fk(x, y), fk+1(x, y);
Difference image binarization threshold is T, and difference image indicates that then the formula of frame differential method is as follows with D (x, y):
Compared with prior art, this programme principle and advantage is as follows:
1. present computer vision field is that deep learning based on neural network occupies dominant position, advantage
Used in large sample can play it in the case where big-sample data, svm is conventional machines learning algorithm, is suitble to do sample
This, the uncomplicated practical application of model, but in situation complexity when needing great amount of samples, has the shortcomings that certain, for example can consume
Plenty of time cost and memory.Therefore, neural network can get in this pedestrian detection model and preferably imitate compared with svm classifier
Fruit.
2. use direction controllable filter algorithm calculates image pixel gradient, arithmetic speed is improved.
3. not detected with whole image of 64*128 window sliding when carrying out pedestrian detection, using frame difference
Method has first determined moving region, and moving region is carried out to certain scaling, then region is detected, and operation is reduced
Time improves detection efficiency.
Detailed description of the invention
Fig. 1 is a kind of work of image pedestrian's detection method based on improved Hog feature combination neural network of the present invention
Flow chart;
Fig. 2 is pedestrian sample figure;
Fig. 3 is the gradient magnitude figure of Fig. 2;
Fig. 4 is Hog feature extraction flow chart;
Fig. 5 is monitor video pedestrian detection flow chart;
Fig. 6 is effect picture.
Specific embodiment
The present invention is further explained in the light of specific embodiments:
As shown in Figure 1, a kind of image pedestrian's inspection based on improved Hog feature combination neural network described in the present embodiment
Survey method, comprising the following steps:
The study stage:
S1-1: pedestrian sample library is established;The resolution ratio of highway monitoring is 1280*64, according to practical pedestrian in video monitoring
Ratio, sampling this size is 64*128, and 30,000 altogether, half is positive sample, i.e. pedestrian sample, and half is negative sample
This, i.e. non-pedestrian sample, and carry out label, positive sample 1, negative sample 0, therein 70 percent is training sample, hundred
/ tri- ten be test sample;
S1-2: pre-processing the sample in step S1-1, by color image gray processing;
S1-3: to pretreated sample extraction Hog feature vector;
S1-4: after extracting Hog feature vector, establishing neural network, be trained to the Hog feature vector of extraction, obtains
Network weight parameter;
Test phase:
S2-1: pretreatment identical with the study stage is carried out to picture sample to be detected;
S2-2: the Hog feature vector by the step S2-1 pretreated picture obtained is calculated;
S2-3: the feature vector that step S2-2 is obtained tests the neural network model after training, calculates network
The accuracy of model enters the application stage if meeting the requirements, and otherwise, adjusts network structure, and return step by actual result
S1-3;
Application stage:
S3-1: the region of video moving object is extracted using frame differential method, is zoomed in and out according to the size in region, region
If being less than fixed sliding window, region is amplified to window size, extract the Hog feature neural network in region to determine whether
For pedestrian.
As shown in figure 4, extracting Hog feature vector, specific step is as follows in above-mentioned steps S1-3 and step S2-2:
A1: use direction controllable filter calculates the gradient of each pixel:
Using the constituted controllable filter of Gauss second dervative, form is as follows:
Wherein,Respectively Gaussian function corresponding direction second dervative,
Expression is as follows:
Coefficient of correspondence is divided into:
The Steerable filter in tectonic level direction and vertical direction, is denoted as F respectively0, Fπ/2, calculating gray level image (x,
Y) the gradient value G of place's pixel both horizontally and verticallyX(x, y), Gy(x, y):
Gx(x, y)=F0*I;
Gy(x, y)=Fπ/2*I;
Wherein, I is gray level image;
Then direction and the amplitude of gradient are calculated separately:
Referring to pedestrian sample Fig. 2 and its gradient magnitude Fig. 3, it can be seen that the edge contour of pedestrian is mentioned well
It takes, edge is more obvious, and illustrates that its gradient value is bigger, and the specific gravity for participating in hereafter statistics with histogram is bigger, and the feature of pedestrian is just
It is more prominent;
A2: building cell factory lattice simultaneously count its histogram of gradients;
Cell factory (cell) is that the basic unit for calculating histogram and the grid of a fixed size, the present embodiment make
It is the region of 8*8 pixel.Histogram is used to count the gradient direction of pixel in cell factory;
The range of gradient direction is 0~180 degree, is divided into nine parts, i.e. histogram has nine sections, each interval range
It is 20 degree;
Nearest Neighbor with Weighted Voting is taken in histogram ballot, i.e., the gradient magnitude of each pixel obtains one nine dimension as ballot weight
Feature vector;
A3: the feature vector of all cell factories of block is together in series and is normalized, obtains the Hog of the block
Feature;
A4: the feature vector of whole blocks is together in series, and obtains the Hog Feature Descriptor of sample;
Among the above, the size of sample is 64*128, and the block that the direction x is calculated according to the following formula has 7 pieces, and the direction y has 15
Block, a block have 4 cell factories, and a cell factory has 9 features, obtain the feature vector of one 3780 dimension;
N=(W-F)/S+1;
Wherein, N is the block counts in the direction x or the direction y, and W is that the line number of image or columns (calculate the direction x columns, meter
Calculate the direction y line number), F is the line number an of block or columns (because being square, line number and columns are equal), S are cell list
The line number or columns of member (because being square, line number and columns are equal).
And the neural network that above-mentioned steps S1-4 is established is three layers of BP neural network, including input layer, hidden layer, output
Layer;Because hog feature has 3780 dimensions, the number of input layer is 3780, and hidden layer is 4096 neurons, this
It is an empirical value, structure is adjusted according to test result when necessary, due to being two classification problems, so the mind of output layer
It is 2 through first number.The process of neural network is passed through in communication process first by hog feature by input layer propagated forward
Hidden layer, if desired output is not achieved in output layer, carries out back-propagation process until output layer, then obtains before
Prediction error further adjust the weight and threshold value of network, thus make the prediction of neural network export constantly approach it is desired
Output trains network, constantly to reach higher precision of prediction.
The training process of BP neural network including the following steps:
(1) network is initialized;BP neural network first determines the input layer number n of network according to sample data, implies
Node layer number l, output layer number of nodes m input layer number are exactly the dimension of input data, and output layer number of nodes is exactly that target is defeated
The number of none fixation of dimension node in hidden layer out, looks for optimal node in hidden layer using trial and error procedure.Also want
Initialize the connection weight w of input layer and hidden layerij, hidden layer and output layer weight wjk, hidden layer threshold value a, output layer threshold
Value b gives the transmission function f (x) between learning rate and neuron;
(2) hidden layer output calculates, and the calculation formula that hidden layer exports H is as follows:
F is activation primitive, is the guarantee of Neural Network Based Nonlinear, and the activation primitive that the present embodiment uses is sigmoid letter
Number, formula are as follows:
(3) output layer is calculated, H, connection weight w are exported according to hidden layerjkWith threshold value b, it is as follows to calculate 0 formula of output:
(4) error is calculated, loss function is as follows:
J=log (sigmoid (OK));
(5) weight is updated, formula is as follows, and η is learning rate:
After the above-mentioned weight parameter for obtaining neural network, Neural Network Data model can be used to carry out pedestrian detection.Cause
It is to traverse whole image using 64*128 window for traditional pedestrian detection method, hog feature is extracted, with classifier to judge
Whether the image-region of traversal is pedestrian, and time-consuming for this method, influences whether real-time, and the present embodiment first uses frame differential method
It determines the region of movement, if the window of movement is less than 64*128, region is amplified to this window size, is then determining
It is traversed in region and obtains hog feature, carried out classification with neural network and judge, the time of operation can be improved.
Frame differential method is by two frame adjacent in video flowing or to be separated by the two images pixel values of a few frame images and subtract each other, and right
Image after subtracting each other carries out thresholding to extract the moving region in image;If the frame number for subtracting each other two field pictures is respectively kth frame,
(k+1) frame, then its frame image is respectively as follows: fk(x, y), fk+1(x, y);
Difference image binarization threshold is T, and difference image indicates that then the formula of frame differential method is as follows with D (x, y):
As shown in figure 5, detailed process is as follows by step S3-1:
B1: video source is obtained;
B2: moving region is determined using frame differential method;
B3: it is slided using fixed window;
B4: judging whether window has target, if so, then storing target area coordinates;If no, return step B3;
B5: judging whether coverage motion region, if so, terminating;If it is not, then return step B3.
Effect picture as shown in FIG. 6 finally can be obtained.
The present embodiment use direction controllable filter algorithm calculates image pixel gradient, improves arithmetic speed.In addition,
When carrying out pedestrian detection, is not detected with whole image of 64*128 window sliding, first determined with frame difference method
Moving region, and moving region is carried out to certain scaling, then region is detected, reduce the time of operation, improves
Detection efficiency.
The examples of implementation of the above are only the preferred embodiments of the invention, and implementation model of the invention is not limited with this
It encloses, therefore all shapes according to the present invention, changes made by principle, should all be included within the scope of protection of the present invention.
Claims (6)
1. a kind of image pedestrian's detection method based on improved Hog feature combination neural network, which is characterized in that including following
Several stages:
The study stage:
S1-1: establishing pedestrian sample library, the non-pedestrian sample of pedestrian sample and half including half;
S1-2: pre-processing the sample in step S1-1, by color image gray processing;
S1-3: to pretreated sample extraction Hog feature vector;
S1-4: establishing neural network, is trained to the Hog feature vector of extraction, obtains network weight parameter;
Test phase:
S2-1: pretreatment identical with the study stage is carried out to picture sample to be detected;
S2-2: the Hog feature vector by the step S2-1 pretreated picture obtained is calculated;
S2-3: the feature vector that step S2-2 is obtained tests the neural network model after training, calculates network model
Accuracy enter the application stage if meeting the requirements, otherwise, by actual result adjustment network structure, and return step S1-
3;
Application stage:
S3-1: the region of video moving object is extracted using frame differential method, is zoomed in and out according to the size in region, if region is small
In fixed sliding window, region is amplified to window size, extracts the Hog feature neural network in region to determine whether for row
People.
2. a kind of image pedestrian's detection method based on improved Hog feature combination neural network according to claim 1,
It is characterized in that, extracting Hog feature vector, specific step is as follows in the step S1-3 and step S2-2:
A1: use direction controllable filter calculates the gradient of each pixel;
A2: building cell factory lattice simultaneously count its histogram of gradients;
A3: being together in series the feature vector of all cell factories of block and be normalized, and the Hog for obtaining the block is special
Sign;
A4: the feature vector of whole blocks is together in series, and obtains the Hog Feature Descriptor of sample.
3. a kind of image pedestrian's detection method based on improved Hog feature combination neural network according to claim 2,
It is characterized in that, the step A1 use direction controllable filter calculates the gradient of each pixel, specific step is as follows:
Using the constituted controllable filter of Gauss second dervative, form is as follows:
Wherein,Respectively Gaussian function corresponding direction second dervative, specifically
Expression formula is as follows:
Coefficient of correspondence is divided into:
The Steerable filter in tectonic level direction and vertical direction, is denoted as F respectively0, Fπ/2, calculate at gray level image (x, y)
The gradient value G of pixel both horizontally and verticallyX(x, y), Gy(x, y):
Gx(x, y)=F0*I;
Gy(x, y)=Fπ/2*I;
Wherein, I is gray level image;
Then direction and the amplitude of gradient are calculated separately:
4. a kind of image pedestrian's detection method based on improved Hog feature combination neural network according to claim 2,
It is characterized in that,
In the step A2, the range of gradient direction is 0~180 degree, is divided into nine parts, i.e. histogram has nine sections, often
A interval range is 20 degree;
Nearest Neighbor with Weighted Voting is taken in histogram ballot, i.e., the gradient magnitude of each pixel obtains the spy of one nine dimension as ballot weight
Levy vector.
5. a kind of image pedestrian's detection method based on improved Hog feature combination neural network according to claim 1,
It is characterized in that, detailed process is as follows by the step S3-1:
B1: video source is obtained;
B2: moving region is determined using frame differential method;
B3: it is slided using fixed window;
B4: judging whether window has target, if so, then storing target area coordinates;If no, return step B3;
B5: judging whether coverage motion region, if so, terminating;If it is not, then return step B3.
6. a kind of image pedestrian's detection method based on improved Hog feature combination neural network according to claim 6,
It is characterized in that, subtract each other two frame adjacent in video flowing or the two images pixel value for being separated by several frame images in the step B2,
And thresholding is carried out to the image after subtracting each other to extract the moving region in image;If the frame number for subtracting each other two field pictures is respectively
K frame, (k+1) frame, then its frame image is respectively as follows: fk(x, y), fk+1(x, y);
Difference image binarization threshold is T, and difference image indicates that then the formula of frame differential method is as follows with D (x, y):
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910351442.3A CN110232314A (en) | 2019-04-28 | 2019-04-28 | A kind of image pedestrian's detection method based on improved Hog feature combination neural network |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201910351442.3A CN110232314A (en) | 2019-04-28 | 2019-04-28 | A kind of image pedestrian's detection method based on improved Hog feature combination neural network |
Publications (1)
Publication Number | Publication Date |
---|---|
CN110232314A true CN110232314A (en) | 2019-09-13 |
Family
ID=67860963
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201910351442.3A Pending CN110232314A (en) | 2019-04-28 | 2019-04-28 | A kind of image pedestrian's detection method based on improved Hog feature combination neural network |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN110232314A (en) |
Cited By (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112488039A (en) * | 2020-12-15 | 2021-03-12 | 哈尔滨市科佳通用机电股份有限公司 | Machine learning-based method for detecting falling fault of hook tail frame supporting plate of railway wagon |
CN113163110A (en) * | 2021-03-05 | 2021-07-23 | 北京宙心科技有限公司 | People stream density analysis system and analysis method |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101827204A (en) * | 2010-04-19 | 2010-09-08 | 成都索贝数码科技股份有限公司 | Method and system for detecting moving object |
CN104166841A (en) * | 2014-07-24 | 2014-11-26 | 浙江大学 | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network |
CN104899559A (en) * | 2015-05-25 | 2015-09-09 | 江苏大学 | Rapid pedestrian detection method based on video monitoring |
CN105321187A (en) * | 2014-07-22 | 2016-02-10 | 无锡慧眼电子科技有限公司 | Pedestrian counting method based on head detection |
CN106778504A (en) * | 2016-11-21 | 2017-05-31 | 南宁市浩发科技有限公司 | A kind of pedestrian detection method |
-
2019
- 2019-04-28 CN CN201910351442.3A patent/CN110232314A/en active Pending
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN101827204A (en) * | 2010-04-19 | 2010-09-08 | 成都索贝数码科技股份有限公司 | Method and system for detecting moving object |
CN105321187A (en) * | 2014-07-22 | 2016-02-10 | 无锡慧眼电子科技有限公司 | Pedestrian counting method based on head detection |
CN104166841A (en) * | 2014-07-24 | 2014-11-26 | 浙江大学 | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network |
CN104899559A (en) * | 2015-05-25 | 2015-09-09 | 江苏大学 | Rapid pedestrian detection method based on video monitoring |
CN106778504A (en) * | 2016-11-21 | 2017-05-31 | 南宁市浩发科技有限公司 | A kind of pedestrian detection method |
Non-Patent Citations (2)
Title |
---|
FAHIM AHMED等: "HOG and Gabor Filter Based Pedestrian Detection using Convolutional Neural Networks", 《2019 IEEE,2019 INTERNATIONAL CONFERENCE ON ELECTRICAL, COMPUTER AND COMMUNICATION ENGINEERING (ECCE)》 * |
耿庆田等: "基于改进HOG特征提取的车型识别算法", 《中国光学》 * |
Cited By (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112488039A (en) * | 2020-12-15 | 2021-03-12 | 哈尔滨市科佳通用机电股份有限公司 | Machine learning-based method for detecting falling fault of hook tail frame supporting plate of railway wagon |
CN113163110A (en) * | 2021-03-05 | 2021-07-23 | 北京宙心科技有限公司 | People stream density analysis system and analysis method |
CN113163110B (en) * | 2021-03-05 | 2022-04-08 | 北京宙心科技有限公司 | People stream density analysis system and analysis method |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN108615226B (en) | Image defogging method based on generation type countermeasure network | |
CN111784633B (en) | Insulator defect automatic detection algorithm for electric power inspection video | |
CN109859171A (en) | A kind of flooring defect automatic testing method based on computer vision and deep learning | |
CN108229338A (en) | A kind of video behavior recognition methods based on depth convolution feature | |
CN107194559A (en) | A kind of work stream recognition method based on Three dimensional convolution neutral net | |
CN110580472B (en) | Video foreground detection method based on full convolution network and conditional countermeasure network | |
CN108280856A (en) | The unknown object that network model is inputted based on mixed information captures position and orientation estimation method | |
CN109670430A (en) | A kind of face vivo identification method of the multiple Classifiers Combination based on deep learning | |
CN110135282B (en) | Examinee return plagiarism cheating detection method based on deep convolutional neural network model | |
CN109918971A (en) | Number detection method and device in monitor video | |
CN108090403A (en) | Face dynamic identification method and system based on 3D convolutional neural network | |
CN110334660A (en) | A kind of forest fire monitoring method based on machine vision under the conditions of greasy weather | |
CN107680116A (en) | A kind of method for monitoring moving object in video sequences | |
CN109241830A (en) | It listens to the teacher method for detecting abnormality in the classroom for generating confrontation network based on illumination | |
CN103489012B (en) | Crowd density detecting method and system based on support vector machine | |
CN108921215A (en) | A kind of Smoke Detection based on local extremum Symbiotic Model and energy spectrometer | |
CN112464844A (en) | Human behavior and action recognition method based on deep learning and moving target detection | |
Janku et al. | Fire detection in video stream by using simple artificial neural network | |
CN106570885A (en) | Background modeling method based on brightness and texture fusion threshold value | |
Liang et al. | Methods of moving target detection and behavior recognition in intelligent vision monitoring. | |
Zhu et al. | Fast detection of moving object based on improved frame-difference method | |
Su et al. | A new local-main-gradient-orientation HOG and contour differences based algorithm for object classification | |
CN110232314A (en) | A kind of image pedestrian's detection method based on improved Hog feature combination neural network | |
CN105930793A (en) | Human body detection method based on SAE characteristic visual learning | |
CN106650824B (en) | Moving object classification method based on support vector machines |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20190913 |
|
RJ01 | Rejection of invention patent application after publication |