CN107038416A - A kind of pedestrian detection method based on bianry image modified HOG features - Google Patents
A kind of pedestrian detection method based on bianry image modified HOG features Download PDFInfo
- Publication number
- CN107038416A CN107038416A CN201710140483.9A CN201710140483A CN107038416A CN 107038416 A CN107038416 A CN 107038416A CN 201710140483 A CN201710140483 A CN 201710140483A CN 107038416 A CN107038416 A CN 107038416A
- Authority
- CN
- China
- Prior art keywords
- pixel
- image
- gradient
- row
- pedestrian detection
- 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
- 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
-
- 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
- G06V10/28—Quantising the image, e.g. histogram thresholding for discrimination between background and foreground patterns
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/50—Extraction of image or video features by performing operations within image blocks; by using histograms, e.g. histogram of oriented gradients [HoG]; by summing image-intensity values; Projection analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Multimedia (AREA)
- Theoretical Computer Science (AREA)
- Human Computer Interaction (AREA)
- Image Analysis (AREA)
Abstract
The invention discloses a kind of pedestrian detection method based on bianry image modified HOG features, comprise the following steps：Set up pedestrian's training sample database；Binary conversion treatment is carried out to sample image；Modified HOG characteristic vectors are extracted to binary image；Gaussian normalization is carried out to modified HOG characteristic vectors, using positive and negative sample training, parameters in SVM models is obtained, sets up Linear SVM model；Video frame images to be detected are pre-processed, binary image is obtained；Calculating obtains current modified HOG characteristic vectors；By modified HOG characteristic vector input linear SVM models, if model output is determined as positive sample, target is detected, target location is exported；Detect whether each window of traversal target occurs by way of window is traveled through.The present invention can effectively solve the problem that the shortcoming that memory consumption is big, detection speed is slow during pedestrian detection.
Description
Technical field
It is more particularly to a kind of special based on bianry image modified HOG the present invention relates to computer vision pedestrian detection field
The pedestrian detection method levied.
Background technology
At present, pedestrian detection method is broadly divided into two classes：Method based on template matches and the method based on statistical learning.
Pedestrian detection algorithm wherein based on statistical learning is because various performance of its statistic algorithm meets various application occasions and progressively obtains
Extensive use is arrived.
In the method based on statistical learning, gradient orientation histogram (Histogram of oriented gradient,
HOG) it is characterized in a kind of very effective Gradient Features, it has very under illumination variation and color change to image pedestrian's detection
Good robustness.This method divides the image into small connected region first, is defined as cell factory, by gathering in cell factory
The gradient of each pixel or the direction histogram at edge, then by set of histograms constitutive characteristic describer altogether.Obtain local
After histogram, the local histogram that need to count cell factory carries out contrast normalization in the bigger scope of image.One
As, the Kuai Nei local histograms feature being made up of four cell factories is calculated, 4 cell factories in block are then distinguished into normalizing
Change.Normalized feature causes tested altimetric image to obtain more preferable robustness to illumination variation and shade.Four after normalization
Local histogram's feature cascade can obtain a HOG characteristic vector.HOG features are in the object inspection based on shape such as pedestrian detection
Very well, follow-up many algorithm of target detection are all extensions on this basis to effect in survey.
The pedestrian detection in intelligent security guard field remains some problems at this stage：In real-time detection field, in real time
Property require it is very high, feature detection operand increase cause real-time to be deteriorated；In video pedestrian detection, the golden word of gray level image
Tower model needs to take very big memory source.Therefore, seeking a kind of pedestrian detection method for overcoming above mentioned problem, with important
Research Significance and practical value.
The content of the invention
It is an object of the invention to overcome the shortcoming and deficiency of prior art there is provided one kind to be based on bianry image modified
The pedestrian detection method of HOG features, can effectively solve the problem that the shortcoming that memory consumption is big, detection speed is slow during pedestrian detection.
The purpose of the present invention is realized by following technical scheme：A kind of row based on bianry image modified HOG features
People's detection method, comprises the following steps：
The study stage：
S1, set up pedestrian's training sample database；
S2, using based on row block pixel local auto-adaptive binarization method to sample image carry out binary conversion treatment, obtain
Binary image；
S3, to binary image extract modified HOG characteristic vectors；
S4, Gaussian normalization is carried out to modified HOG characteristic vectors, using positive and negative sample training, obtain in SVM models each
Individual parameter, sets up Linear SVM model；
Decision phase：
S5, video frame images to be detected are pre-processed, obtain binary image；
S6, calculating obtain current modified HOG characteristic vectors；
S7, the Linear SVM model for obtaining modified HOG characteristic vector input steps S4, if model output is determined as
Positive sample, then detect target, exports target location；Detect whether each window of traversal occurs by way of window is traveled through
Target.
It is preferred that, the selection of pedestrian's training sample database follows following two rules in step S1：
Rule 1：Pedestrian's negative sample is with pedestrian's positive sample quantitative proportion 10:1；
Rule 2：Using the SVM of first time training, difficult negative sample, raising are used as by the negative sample of flase drop in detection negative sample
The ratio of these difficult negative samples, so as to further improve the degree of accuracy for the model set up.
It is preferred that, it is specially based on row block pixel local auto-adaptive binarization method in step S2：Read and regard in FPGA
Frequency according to when, it is only necessary to two row image stream informations are preserved in RAM；After being updated using the i-th row gray-scale map gray value and the i-th -1 row
Row vector image intensity value update the gray value of the i-th row pixel together, the updated value of the 1st row pixel is original gray value sheet
Body, such as following formula：
Wherein preY (i, j) refers to the row vector image intensity value after updating, and cur (i, j) refers to gray-scale map gray value；
The binary-state threshold of each pixel is defined below：Take 1*w row blocks big row vector respective pixel after renewal
Small pixel region, tries to achieve the average E (i, j) in the region, the pixel according to required by being determined pixel average E (i, j) in row block
The threshold value of binaryzation, is that can obtain binary image using equation below：
Wherein I (i, j) is the gray value of binary image, and σ is threshold coefficient.
It is preferred that, step S3's comprises the concrete steps that：The gradient magnitude and angle of each pixel are extracted using gradient template,
Statistics with histogram is carried out further according to gradient direction, modified HOG characteristic vectors are obtained.
Specifically, calculating the gradient information of bianry image regional area using 5*5 gradient template, formwork calculation is being carried out
When, by central pixel point and h (0,0) correspondences, then by the pixel h corresponding with template of the pixel around central pixel point
(a, b) is multiplied；Equation below is the universal method for calculating gradient：
Wherein f is binary image, and (x, y) is pixel, for 5*5 template, a and b may value be ± 2,
± 1 and 0；
Utilize hxDirection gradient formwork calculation obtains x direction gradients gx(x, y), utilizes hyDirection gradient formwork calculation obtains y
Direction gradient gy(x,y)；Calculating gradient magnitude and the formula of angle are respectively：
It is that can obtain modified HOG spies to carry out statistics with histogram according to gradient direction further according to each pixel gradient magnitude
Levy vector.
Further, carry out taking Direction interval not decile mode during statistics with histogram, 0~π intervals are pressedArea
Between be divided into an interval,It is interval according toBe divided at equal intervals 5 intervals,It is divided into an interval, totally 7
Individual interval, therefore histogrammic passage totally 7 passages.
Specifically, selection image detection window size 64*128, block size is 16*16, cell size 8*8, block, which is offset, is
8*8, number of blocks 105, each unit includes 8*8 pixel, utilizes binarized pixel gradient of the different directions angle to 8*8 pixel
Information is voted；Nearest Neighbor with Weighted Voting is taken in histogram ballot, i.e., the gradient magnitude of each pixel obtains one as ballot weight
The characteristic vector of multidimensional.
It is preferred that, in step S4 in characteristic vector each element carry out Gaussian normalization processing, so as to get point
The span 99% of value will fall between [0,1], obtain normalized modified HOG characteristic vectors；Change normalized
Enter type HOG characteristic vectors input support vector cassification model to be trained, so as to obtain the pedestrian detection of modified HOG features
Model；Set up Linear SVM model：F (x)=wTX+b, x are edge gradient feature operator α, wTIt is exactly to be instructed by positive negative sample with b
The SVM parameters got.
It is preferred that, the specific steps pre-processed in step S5：Two field picture is extracted to input video, RGB image is converted into
Gray-scale map, after handling after filtering, at based on row block pixel local auto-adaptive binarization method to sample image binaryzation
Reason, obtains binary image.
It is preferred that, pyramid model is set up in step S6 to area-of-interest binary image, level0 layers refer to original
Beginning gray-scale map, level1 refer to original gradation figure reduce 0.95 times after image, the like until zooming to image resolution
Rate is not more than 64*64, and every layer of zoom ratio is all 0.95；To every tomographic image computed improved type HOG characteristic vectors and carry out successively
Gaussian normalization processing, obtains normalized modified HOG characteristic vectors.
The present invention compared with prior art, has the following advantages that and beneficial effect：
1st, the present invention does pedestrian detection using bianry image, relative to 8bit gray-scale maps, can greatly improve detection system
The real-time and reduction memory consumption of system.
2nd, the present invention is using row block pixel local auto-adaptive binarization method is based on, from the multi-line images fluxion of memory storage
According to two row image data streams are reduced to, memory space consumption is greatly reduced.
3rd, the present invention is used as the characteristic vector of bianry image using modified HOG characteristic vectors, greatly remains two-value
The HOG characteristic angle information of image, is that follow-up statistics with histogram remains mass efficient gradient information, improves pedestrian detection
Accuracy rate.
Brief description of the drawings
Fig. 1 is the flow chart of the present embodiment method；
Fig. 2 is the modified HOG feature extraction flow charts of bianry image；
Fig. 3 is gradient template figure：Fig. 3 (a) is X-direction gradient template；Fig. 3 (b) is Y-direction gradient template；
Fig. 4 is the interval division figure of modified HOG characteristic directions.
Embodiment
With reference to embodiment and accompanying drawing, the present invention is described in further detail, but embodiments of the present invention are not limited
In this.
A kind of pedestrian detection method based on bianry image modified HOG features, such as Fig. 1 comprises the following steps：
The study stage：
S1, set up pedestrian's training sample database.
The selection of pedestrian's training sample database follows following two rules：
Rule 1：Pedestrian's negative sample is with pedestrian's positive sample quantitative proportion 10:1；
Rule 2：Using the SVM of first time training, difficult negative sample, raising are used as by the negative sample of flase drop in detection negative sample
The ratio of these difficult negative samples, so as to further improve the degree of accuracy for the model set up.
S2, to sample image carry out binary conversion treatment.
Traditional local auto-adaptive binarization method is applied in FPGA, it is necessary to reserve multirow (n in line of input image stream
OK) the memory space of image stream.Different from directly retaining n row gray level image pixels, the present embodiment utilizes the i-th row gray-scale map gray scale
Row vector image intensity value after value and the i-th -1 row update comes together to update the updated value of the i-th row pixel (the 1st row pixel is more
It is new to be worth for original gray value in itself), such as following formula：
Wherein preY (i, j) refers to the row vector image intensity value after updating, and cur (i, j) refers to gray-scale map gray value.Reading
During video data, it is only necessary to preserve two row image stream informations in RAM, the method is referred to as being based on row block in the present embodiment
Pixel local auto-adaptive binarization method.Two are carried out to sample image using based on row block pixel local auto-adaptive binarization method
Value is handled, and the binary-state threshold of each pixel is defined below：Take 1*w row blocks big row vector respective pixel after renewal
Small pixel region, tries to achieve the average E (i, j) in the region, the pixel according to required by being determined pixel average E (i, j) in row block
The threshold value of binaryzation, is that can obtain binary image using equation below：
Wherein I (i, j) is the gray value of binary image, and σ is threshold coefficient.
S3, to binary image extract modified HOG features, such as Fig. 2.
The gradient magnitude and angle of each pixel are extracted using gradient template, column hisgram system is entered further according to gradient direction
Meter, obtains modified HOG characteristic vectors.Gradient magnitude and angular dimension are calculated, gradient calculation, side are completed using gradient template
Edge Gradient Features are very sensitive for gradient template.
The gradient information of bianry image regional area, statistical gradient direction histogram structure are calculated using 5*5 gradient template
Into modified HOG features.Replace traditional First-order Gradient formwork calculation gradient magnitude and angle big with 5*5 gradient template first
Small, the gradient magnitude direction that conventional first order gradient template calculates bianry image only has four angle directions, uses the gradient of 5*5 ranks
The gradient direction of template generation can retain Gradient direction information so that the angle character of HOG characteristic vectors obtains very big
It is abundant, it is to avoid the low phenomenon of the excessively sparse verification and measurement ratio brought of angle direction.
Here take 5*5 gradient template as shown in figure 3, carry out formwork calculation when, by central pixel point and h (0,0) it is right
Should, then the pixel h (a, b) corresponding with template of the pixel around central pixel point is multiplied.Equation below is terraced to calculate
The universal method of degree：
Wherein f is binary image, and (x, y) is pixel, for 5*5 template, a and b may value be ± 2,
± 1 and 0.
Utilize hxDirection gradient formwork calculation obtains x direction gradients gx(x, y), utilizes hyDirection gradient formwork calculation obtains y
Direction gradient gy(x,y).Calculating gradient magnitude and the formula of angle are respectively：
Statistics with histogram is carried out according to gradient direction further according to each pixel gradient magnitude, because bianry image HOG is special
The gradient angle levied more is concentrated in 0 and π angular intervals, therefore takes Direction interval not decile mode, such as Fig. 4, and 0~π is interval
PressInterval be divided into an interval,It is interval according toBe divided at equal intervals 5 intervals,It is divided into
One interval, totally 7 intervals (bin), therefore histogrammic passage totally 7 passages, histogrammic statistical result is modified
HOG characteristic vectors.
Image detection window size 64*128 is selected, block size is 16*16, and cell size 8*8, block skew is 8*8, block number
Amount 105, each unit includes 8*8 pixel, and the binarized pixel gradient information of 8*8 pixel is carried out using 7 orientation angles
Ballot.Particularly point out, Nearest Neighbor with Weighted Voting is taken in histogram ballot, i.e., the gradient magnitude of each pixel obtains one as ballot weight
The characteristic vector of individual 7 dimension.Each detection block is to have 2940 dimension modified HOG characteristic vectors.
S4, Gaussian normalization is carried out to modified HOG characteristic vectors, using positive and negative sample training, obtain in SVM models each
Individual parameter, sets up Linear SVM model.
To in characteristic vector each element carry out Gaussian normalization processing, so as to get component value span
99% will fall between [0,1], obtain normalized modified HOG characteristic vectors.By normalized modified HOG characteristic vectors
Input support vector cassification model is trained, so as to obtain the pedestrian detection model of modified HOG features.
Set up Linear SVM model：F (x)=wTX+b, x are edge gradient feature operator α, wTIt is exactly to pass through positive negative sample with b
Train obtained SVM parameters.
Decision phase：
S5, video frame images to be detected are pre-processed, obtain binary image.
In the present embodiment, in pedestrian's detection field, the specific steps of pretreatment：Two field picture is extracted to input video, will
RGB image is converted into gray-scale map, after handling after filtering, based on row block pixel local auto-adaptive binaryzation to sample image two-value
Change is handled, and obtains binary image.
S6, calculating obtain current modified HOG characteristic vectors.
Different from zooming in and out detection to 8bit gray-scale maps in conventional video detection, the present embodiment is by input video two field picture
Binaryzation obtains 1bit binary image, sets up the pyramid zoom model of binary image.In being consumed in pedestrian detection
It is at most exactly, to the pyramid scaling of image and detection, 8bit gray level images to be changed into 1bit two-values here to deposit maximum, time-consuming
Figure, greatly reduces memory consumption, accelerates detection speed.
Pyramid model is set up to binary image, level0 layers refer to original gradation figure, and level1 refers to original
Gray-scale map reduce 0.95 times after image, the like until zooming to image resolution ratio no more than 64*64, every layer of zoom ratio
All it is 0.95.
To every tomographic image computed improved type HOG characteristic vectors and Gaussian normalization processing is carried out successively, obtain normalized
Modified HOG characteristic vectors.
S7, the Linear SVM model that step S4 is obtained is inputted, if model output is determined as positive sample, detected
Target, exports target location.
According to f (x)=w in Linear SVM modelTThe model parameter that x+b study is obtained, normalized modified HOG is special
Vectorial importing is levied, according to the rule for more than 0 being positive sample, target location is exported.
To area-of-interest, using pyramid model, detected by way of window is traveled through traversal each window whether
There is target.
Above-described embodiment is preferably embodiment, but embodiments of the present invention are not by above-described embodiment of the invention
Limitation, other any Spirit Essences without departing from the present invention and the change made under principle, modification, replacement, combine, simplification,
Equivalent substitute mode is should be, is included within protection scope of the present invention.
Claims (10)
1. a kind of pedestrian detection method based on bianry image modified HOG features, it is characterised in that comprise the following steps：
The study stage：
S1, set up pedestrian's training sample database；
S2, using based on row block pixel local auto-adaptive binarization method to sample image carry out binary conversion treatment, obtain two-value
Change image；
S3, to binary image extract modified HOG characteristic vectors；
S4, Gaussian normalization is carried out to modified HOG characteristic vectors, using positive and negative sample training, obtain in SVM models each and join
Number, sets up Linear SVM model；
Decision phase：
S5, video frame images to be detected are pre-processed, obtain binary image；
S6, calculating obtain current modified HOG characteristic vectors；
S7, the Linear SVM model for obtaining modified HOG characteristic vector input steps S4, if model output is determined as positive sample
This, then detect target, exports target location；Detect whether each window of traversal mesh occurs by way of window is traveled through
Mark.
2. pedestrian detection method according to claim 1, it is characterised in that the selection of pedestrian's training sample database in step S1
Follow following two rules：
Rule 1：Pedestrian's negative sample is with pedestrian's positive sample quantitative proportion 10:1；
Rule 2：Using the SVM of first time training, detect in negative sample by the negative sample of flase drop as difficult negative sample, improve these
The ratio of difficult negative sample, so as to further improve the degree of accuracy for the model set up.
3. pedestrian detection method according to claim 1, it is characterised in that locally adaptive based on row block pixel in step S2
The binarization method is answered to be specially：In FPGA during reading video data, it is only necessary to two row image stream informations are preserved in RAM；Profit
Row vector image intensity value after being updated with the i-th row gray-scale map gray value and the i-th -1 row updates the gray scale of the i-th row pixel together
Value, the updated value of the 1st row pixel for original gray value in itself, such as following formula：
Wherein preY (i, j) refers to the row vector image intensity value after updating, and cur (i, j) refers to gray-scale map gray value；
The binary-state threshold of each pixel is defined below：1*w row block sizes are taken to the row vector respective pixel after renewal
Pixel region, tries to achieve the average E (i, j) in the region, the pixel two-value according to required by being determined pixel average E (i, j) in row block
The threshold value of change, is that can obtain binary image using equation below：
Wherein I (i, j) is the gray value of binary image, and σ is threshold coefficient.
4. pedestrian detection method according to claim 1, it is characterised in that step S3's comprises the concrete steps that：Utilize gradient
The gradient magnitude and angle of each pixel of template extraction, carry out statistics with histogram further according to gradient direction, obtain modified HOG
Characteristic vector.
5. pedestrian detection method according to claim 1, it is characterised in that calculate binary map using 5*5 gradient template
As the gradient information of regional area, when carrying out formwork calculation, by central pixel point and h (0,0) correspondences, then by center pixel
Pixel pixel h (a, b) corresponding with template around point is multiplied；Equation below is the universal method for calculating gradient：
Wherein f is binary image, and (x, y) is pixel, for 5*5 template, and a and the possible values of b are ± 2, ± 1 Hes
0；
Utilize hxDirection gradient formwork calculation obtains x direction gradients gx(x, y), utilizes hyDirection gradient formwork calculation obtains y directions
Gradient gy(x,y)；Calculating gradient magnitude and the formula of angle are respectively：
Further according to each pixel gradient magnitude according to gradient direction carry out statistics with histogram be can obtain modified HOG features to
Amount.
6. pedestrian detection method according to claim 4, it is characterised in that carry out taking Direction interval during statistics with histogram
Decile mode, not interval by 0~πInterval be divided into an interval,It is interval according toIt is divided into 5 at equal intervals
Individual interval,It is divided into an interval, totally 7 intervals, histogrammic passage totally 7 passages.
7. pedestrian detection method according to claim 4, it is characterised in that selection image detection window size 64*128,
Block size is 16*16, and cell size 8*8, block skew is 8*8, and number of blocks 105, each unit includes 8*8 pixel, using not
Equidirectional angle is voted the binarized pixel gradient information of 8*8 pixel；Nearest Neighbor with Weighted Voting is taken in histogram ballot, i.e., each
The gradient magnitude of pixel obtains the characteristic vector of a multidimensional as ballot weight.
8. pedestrian detection method according to claim 1, it is characterised in that to each in characteristic vector in step S4
Element carries out Gaussian normalization processing, so as to get the span 99% of component value will fall between [0,1], normalized
Modified HOG characteristic vectors；Normalized modified HOG characteristic vectors input support vector cassification model is instructed
Practice, so as to obtain the pedestrian detection model of modified HOG features；Set up Linear SVM model：F (x)=wTX+b, x are edge ladders
Spend feature operator α, wTIt is exactly the SVM parameters that are obtained by positive and negative sample training with b.
9. pedestrian detection method according to claim 1, it is characterised in that the specific steps pre-processed in step S5：It is right
Input video extracts two field picture, and RGB image is converted into gray-scale map, after handling after filtering, using based on row block pixel part
Self-adaption binaryzation method obtains binary image to sample image binary conversion treatment.
10. pedestrian detection method according to claim 1, it is characterised in that to area-of-interest binaryzation in step S6
Image sets up pyramid model, and level0 layers refer to original gradation figure, and level1 refers to reduce 0.95 times to original gradation figure
Image afterwards, the like until zooming to image resolution ratio no more than 64*64, every layer of zoom ratio is all 0.95；It is right successively
Per tomographic image computed improved type HOG characteristic vectors simultaneously carry out Gaussian normalization processing, obtain normalized modified HOG features to
Amount.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710140483.9A CN107038416B (en) | 2017-03-10 | 2017-03-10 | Pedestrian detection method based on binary image improved HOG characteristics |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710140483.9A CN107038416B (en) | 2017-03-10 | 2017-03-10 | Pedestrian detection method based on binary image improved HOG characteristics |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107038416A true CN107038416A (en) | 2017-08-11 |
| CN107038416B CN107038416B (en) | 2020-02-18 |
Family
ID=59534454
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710140483.9A Active CN107038416B (en) | 2017-03-10 | 2017-03-10 | Pedestrian detection method based on binary image improved HOG characteristics |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107038416B (en) |
Cited By (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679528A (en) * | 2017-11-24 | 2018-02-09 | 广西师范大学 | A kind of pedestrian detection method based on AdaBoost SVM Ensemble Learning Algorithms |
| CN108182442A (en) * | 2017-12-29 | 2018-06-19 | 惠州华阳通用电子有限公司 | A kind of image characteristic extracting method |
| CN108460420A (en) * | 2018-03-13 | 2018-08-28 | 江苏实达迪美数据处理有限公司 | A method of classify to certificate image |
| CN108710909A (en) * | 2018-05-17 | 2018-10-26 | 南京汇川工业视觉技术开发有限公司 | A kind of deformable invariable rotary vanning object counting method |
| CN108875628A (en) * | 2018-06-14 | 2018-11-23 | 攀枝花学院 | pedestrian detection method |
| CN109034125A (en) * | 2018-08-30 | 2018-12-18 | 北京工业大学 | Pedestrian detection method and system based on scene complexity |
| CN110765877A (en) * | 2019-09-20 | 2020-02-07 | 南京理工大学 | Pedestrian detection method and system based on thermal imager and binocular camera |
| CN110866534A (en) * | 2019-08-13 | 2020-03-06 | 广州三木智能科技有限公司 | Far infrared pedestrian training method for gradient amplitude distribution gradient orientation histogram |
| CN111913584A (en) * | 2020-08-19 | 2020-11-10 | 福州大学 | Mouse cursor control method and system based on gesture recognition |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663409A (en) * | 2012-02-28 | 2012-09-12 | 西安电子科技大学 | Pedestrian tracking method based on HOG-LBP |
| CN103177262A (en) * | 2013-02-19 | 2013-06-26 | 山东大学 | FPGA (field programmable gate array) architecture of HOG (histogram of oriented gradient) and SVM (support vector machine) based pedestrian detection system and implementing method of FPGA architecture |
| CN103440035A (en) * | 2013-08-20 | 2013-12-11 | 华南理工大学 | Gesture recognition system in three-dimensional space and recognition method thereof |
| CN103530638A (en) * | 2013-10-29 | 2014-01-22 | 无锡赛思汇智科技有限公司 | Method for matching pedestrians under multiple cameras |
| CN103745226A (en) * | 2013-12-31 | 2014-04-23 | 国家电网公司 | Dressing safety detection method for worker on working site of electric power facility |
-
2017
- 2017-03-10 CN CN201710140483.9A patent/CN107038416B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663409A (en) * | 2012-02-28 | 2012-09-12 | 西安电子科技大学 | Pedestrian tracking method based on HOG-LBP |
| CN103177262A (en) * | 2013-02-19 | 2013-06-26 | 山东大学 | FPGA (field programmable gate array) architecture of HOG (histogram of oriented gradient) and SVM (support vector machine) based pedestrian detection system and implementing method of FPGA architecture |
| CN103440035A (en) * | 2013-08-20 | 2013-12-11 | 华南理工大学 | Gesture recognition system in three-dimensional space and recognition method thereof |
| CN103530638A (en) * | 2013-10-29 | 2014-01-22 | 无锡赛思汇智科技有限公司 | Method for matching pedestrians under multiple cameras |
| CN103745226A (en) * | 2013-12-31 | 2014-04-23 | 国家电网公司 | Dressing safety detection method for worker on working site of electric power facility |
Cited By (13)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107679528A (en) * | 2017-11-24 | 2018-02-09 | 广西师范大学 | A kind of pedestrian detection method based on AdaBoost SVM Ensemble Learning Algorithms |
| CN108182442A (en) * | 2017-12-29 | 2018-06-19 | 惠州华阳通用电子有限公司 | A kind of image characteristic extracting method |
| CN108460420A (en) * | 2018-03-13 | 2018-08-28 | 江苏实达迪美数据处理有限公司 | A method of classify to certificate image |
| CN108710909A (en) * | 2018-05-17 | 2018-10-26 | 南京汇川工业视觉技术开发有限公司 | A kind of deformable invariable rotary vanning object counting method |
| CN108875628A (en) * | 2018-06-14 | 2018-11-23 | 攀枝花学院 | pedestrian detection method |
| CN109034125B (en) * | 2018-08-30 | 2021-12-03 | 北京工业大学 | Pedestrian detection method and system based on scene complexity |
| CN109034125A (en) * | 2018-08-30 | 2018-12-18 | 北京工业大学 | Pedestrian detection method and system based on scene complexity |
| CN110866534A (en) * | 2019-08-13 | 2020-03-06 | 广州三木智能科技有限公司 | Far infrared pedestrian training method for gradient amplitude distribution gradient orientation histogram |
| CN110866534B (en) * | 2019-08-13 | 2023-09-12 | 广州三木智能科技有限公司 | Far infrared pedestrian training method for gradient amplitude distribution gradient orientation histogram |
| CN110765877A (en) * | 2019-09-20 | 2020-02-07 | 南京理工大学 | Pedestrian detection method and system based on thermal imager and binocular camera |
| CN110765877B (en) * | 2019-09-20 | 2022-09-06 | 南京理工大学 | Pedestrian detection method and system based on thermal imager and binocular camera |
| CN111913584A (en) * | 2020-08-19 | 2020-11-10 | 福州大学 | Mouse cursor control method and system based on gesture recognition |
| CN111913584B (en) * | 2020-08-19 | 2022-04-01 | 福州大学 | Mouse cursor control method and system based on gesture recognition |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107038416B (en) | 2020-02-18 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107038416A (en) | A kind of pedestrian detection method based on bianry image modified HOG features | |
| CN101236608B (en) | Human face detection method based on picture geometry | |
| CN103578119B (en) | Target detection method in Codebook dynamic scene based on superpixels | |
| CN108491797A (en) | A kind of vehicle image precise search method based on big data | |
| CN107229929A (en) | A kind of license plate locating method based on R CNN | |
| CN108960198A (en) | A kind of road traffic sign detection and recognition methods based on residual error SSD model | |
| CN107742099A (en) | A kind of crowd density estimation based on full convolutional network, the method for demographics | |
| CN106991666B (en) | A kind of disease geo-radar image recognition methods suitable for more size pictorial informations | |
| CN104850836A (en) | Automatic insect image identification method based on depth convolutional neural network | |
| CN104166841A (en) | Rapid detection identification method for specified pedestrian or vehicle in video monitoring network | |
| CN104992223A (en) | Dense population estimation method based on deep learning | |
| CN107169985A (en) | A kind of moving target detecting method based on symmetrical inter-frame difference and context update | |
| CN107871316B (en) | Automatic X-ray film hand bone interest area extraction method based on deep neural network | |
| CN104915972A (en) | Image processing apparatus, image processing method and program | |
| CN107545571A (en) | A kind of image detecting method and device | |
| CN110084165A (en) | The intelligent recognition and method for early warning of anomalous event under the open scene of power domain based on edge calculations | |
| CN107633226A (en) | A kind of human action Tracking Recognition method and system | |
| CN104636755A (en) | Face beauty evaluation method based on deep learning | |
| CN109978032A (en) | Bridge Crack detection method based on spatial pyramid cavity convolutional network | |
| CN109034184A (en) | A kind of grading ring detection recognition method based on deep learning | |
| CN104978567A (en) | Vehicle detection method based on scenario classification | |
| CN108256462A (en) | A kind of demographic method in market monitor video | |
| CN109002752A (en) | A kind of complicated common scene rapid pedestrian detection method based on deep learning | |
| CN105469111A (en) | Small sample set object classification method on basis of improved MFA and transfer learning | |
| CN108021890A (en) | A kind of high score remote sensing image harbour detection method based on PLSA and BOW |
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 |