CN103279759A - Vehicle front trafficability analyzing method based on convolution nerve network - Google Patents
Vehicle front trafficability analyzing method based on convolution nerve network Download PDFInfo
- Publication number
- CN103279759A CN103279759A CN2013102341260A CN201310234126A CN103279759A CN 103279759 A CN103279759 A CN 103279759A CN 2013102341260 A CN2013102341260 A CN 2013102341260A CN 201310234126 A CN201310234126 A CN 201310234126A CN 103279759 A CN103279759 A CN 103279759A
- Authority
- CN
- China
- Prior art keywords
- image
- pixel
- gray
- layer
- scale value
- 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
Images
Landscapes
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a vehicle front trafficability analyzing method based on a convolution nerve network. The method comprises the following steps: first, a vidicon arranged in the front of a vehicle is used for collecting a large number of actual vehicle traveling environment images; a Gamma rectification function is used for pre-processing the images; training of the convolution nerve network is conducted. According to the method, a nonlinear function superimposed Gamma rectification method is used for pre-processing the images, so that influence of light illumination of strong changes on identification of objects is avoided, and the image resolution ratio is improved. According to the method, a geometry normalization method is used, so that the resolution ratio difference caused by identifying the distance of an object distance vidicon is reduced. The convolution nerve network LeNet-5 adopted in the method can extract implicit expression characteristics with class distinguishing capacity and is simple in extracting process. The LeNet-5 is combined with a local receptive field, weight share and secondary sampling to ensure robustness of simple geometry deformation, reduce training parameters of the network, and simplify the structure of the network.
Description
Technical field
The invention belongs to safe driver assistance and intelligent transport technology field, relate to vehicle front trafficability analytical approach, specially refer to a kind of by camera acquisition vehicle front video image, based on the vehicle front trafficability analytical approach of convolutional neural networks.
Background technology
The analysis of vehicle front trafficability belongs to the environment sensing category of intelligent transportation field outside car, refer to based on advanced meanses such as sensor technology, computer technology or mechanicss of communication the driving safety of institute's acquisition environment be analyzed, find out the potential safety hazard of existence, send prompting and early warning or lay the foundation for the automatic driving vehicle navigation to the driver.At present, based on camera acquisition vehicle front video image information, the research of adopting the visual pattern understanding method to carry out the trafficability analysis mainly contains detection of obstacles, pedestrian detection, vehicle detection, Road Detection, traffic sign detection, topography and geomorphology classification etc.
Trafficability is analyzed related visual pattern understanding method and can be divided into based on the method for rebuilding with based on the method for identifying.Wherein, be based on three-dimensional or 2.5 dimension reconstruction techniques based on the method for rebuilding, make judgement from the angle in space to whether passing through, problems such as the less and real-time difference of serious polysemy, the reconstruction scope that is difficult to avoid three-dimensional reconstruction intrinsic.In the image understanding method based on identification, mainly contain algorithm based on modeling and template matches, general neural network, support vector machine, self-supervisory study, based on the method for statistical learning etc., these methods need be extracted the explicit features of target, the leaching process complexity, easily cause important information to lose, adaptive capacity to environment is poor.
For the strong structured road environment of illumination variation, if directly original image is identified, interfere information is many, the process complexity that explicit features is extracted, furthermore target can cause the difference of resolution apart from difference apart from video camera.In addition, the variation of illumination can influence picture quality, reduces the resolution of image.
Summary of the invention
For solving the problems referred to above that prior art exists, the present invention will propose a kind of vehicle front trafficability analytical approach based on convolutional neural networks, this method can be extracted the implicit features of target, leaching process is simple, avoid reducing the resolution of image, and can reduce the influence of illumination, be applicable to the structured road environment that illumination variation is strong.
Technical scheme of the present invention is: a kind of vehicle front trafficability analytical approach based on convolutional neural networks may further comprise the steps:
A, image acquisition
At first by a large amount of real vehicle running environment images of the camera acquisition that is installed in vehicle front, described image has m * n pixel; Obtain the zone of image bottom 3/5ths as area-of-interest by cutting then; At last images cut is converted into gray level image.
B, image pre-service
B1, the Gamma of method construct that utilizes nonlinear function to superpose correct function, and the gray level image that steps A obtains is corrected, and concrete functional expression is as follows:
G(x)＝1+f
1(x)+f
2(x)+f
3(x) （1）
f
1(x)＝acos(πx/255) （2）
f
2(x)＝(K(x)+b)cosβ+xsinα （3）
K(x)＝ρsin(4πx/255) （4）
α＝arctan(-2b/255) （5）
f
3(x)＝R(x)cos(3πx/255) （6）
R(x)＝c|2x/255-1| （7）
In the formula, x is a certain gray values of pixel points, the Gamma correction value of a certain gray-scale value correspondence of G (x) representative, and a ∈ (0,1) is a weighting coefficient, b represents f
2(x) maximum changing range, ρ are represented the amplitude of K (x), and α represents the deflection angle of K (x), and c represents the amplitude of R (x), and satisfy a+b+c＜1.
Gray-scale value computing formula after Gamma corrects is:
g(x)＝255(x/255)
1/G(x) （8）
In the formula, a certain gray values of pixel points after g (x) representative is corrected through Gamma.
Correct through Gamma, obtain gray level image P.
B2, at gray level image P, change some gray values of pixel points, concrete change method is as follows:
Choose in the image gray-scale value in the image-region except vehicle and road boundary and be 0 pixel, change its gray-scale value into 1, choose in the image gray-scale value in the image-region except vehicle and road boundary and be 255 pixel, change its gray-scale value into 254; Change vehicle region gray values of pixel points in the image into 0, the gray-scale value of road boundary area pixel point changes 255 into, and the image behind the change pixel is gray level image Q.So far, the pixel of gray level image Q comprises three classes: the first kind is that gray-scale value is 0 pixel, represents vehicle; Second class is that gray-scale value is 255 pixel, represents road boundary; The 3rd class is that to remove gray-scale value be pixel outside 0 and 255, represents the road surface.Give corresponding label respectively with above three class pixels, be about to label " 0 " and compose expression " vehicle " to first kind pixel, label " 1 " is composed to the second class pixel, expression " road boundary " composes label " 2 " to the 3rd class pixel, expression " road surface ".At last the label of each pixel among the gray level image Q is composed to corresponding pixel among the gray level image P.
B3, carry out normalized at the size of gray level image P:
B31, along the picture altitude direction, at interval choose different pixel columns, represent with x, measure by actual samples, obtain the pixel wide of the corresponding target of different pixels row x institute and highly;
B32, being that 0～32 image-region is the reference picture zone with pixels tall in the image, is benchmark with pixel wide W and the height H of the target that will identify in this reference picture zone, and the horizontal and vertical cutting scale-up factor of namely establishing this reference picture zone is 1; With W and H remove respectively target on all the other each pixel columns width and the height, obtain two groups of ratios, represent with Y and Z respectively;
B33, at last pixel column x and two groups of ratio Y and Z are carried out match respectively, obtain two matched curves, as follows:
Y＝k
1x+b
1 （9）
Z＝k
2x+b
2 （10）
Wherein, the horizontal cutting scale-up factor of Y representative image, the vertical cutting scale-up factor of Z representative image, a certain pixel column of x representative image, k
1, k
2Represent the slope of two matched curves respectively, b
1, b
2Represent the intercept of two matched curves respectively.
B34, the horizontal and vertical cutting ratio in reference picture zone all is made as 1, is about to the image pattern that the reference picture zone is cut to 32 * 32 pixels.Along with the increase of x, according to the also corresponding increase of horizontal and vertical cutting size of formula (9), (10) acquisition.By cutting, obtain a series of image patterns not of uniform size, the image pattern unification that at last cutting is obtained is scaled the image of 32 * 32 pixels.With the image of 32 * 32 pixels that the obtain training sample as convolutional neural networks.
The training of C, convolutional neural networks
Typical case's convolutional neural networks LeNet-5 forms by 8 layers, and input is that layer is 32 * 32 pixel images; Network layer C1, C3 and C5 represent convolutional layer respectively, and network layer S2 and S4 are time sampling layer, and network layer F5 is full articulamentum, and the neuronic number of output layer is identical with the target classification number that will identify, changes according to actual application environment.A face of every layer network represents a characteristic pattern, the set that the neuron that this characteristic pattern is shared by weights in each layer is formed.The neuron of each layer only is connected with the neuron of a local receptive field of last layer.
The general type of convolutional layer is:
In the formula, { 1,2,3,4,5,6,7,8} represents the number of plies to l ∈, and k is convolution kernel, M
jRepresent the selection of input feature vector figure, b represents biasing.
The general type of inferior sampling layer is:
In the formula, down () expression time sampling function generally is the zone summation to a n * n of last tomographic image, and β represents time weights of sampling layer, and b represents biasing.
According to actual application environment, the number of the output neuron of LeNet-5 to be adjusted, the Pixel Dimensions that adopts step B to obtain then is that 32 * 32 image pattern is trained.By training, when the output valve of convolutional neural networks and the error of expectation value are in the tolerance interval, just obtain can be used for the convolutional neural networks of vehicle front trafficability analysis.
Compared with prior art, effect of the present invention and benefit are:
1, the present invention adopts the Gamma antidote pretreatment image of nonlinear function stack, has avoided the illumination of strong variations to the influence of target identification, has improved image resolution ratio.
2, the present invention has adopted the geometrical normalization method, has reduced the differences in resolution that identification target range video camera distance causes.
3, the convolutional neural networks LeNet-5 of the present invention's employing can extract the implicit features with classification resolution characteristic, and leaching process is simple; LeNet-5 shares and inferior sampling in conjunction with local receptive field, weights, guarantees the robustness to the simple geometry distortion, and has reduced the training parameter of network, has simplified network structure; The neuron number of LeNet-5 output layer can be adjusted according to actual application environment, and adaptive capacity to environment is strong.
Description of drawings
3 in the total accompanying drawing of the present invention, wherein:
Fig. 1 is based on the vehicle front trafficability analytical approach process flow diagram of convolutional neural networks
Fig. 2 convolutional neural networks LeNet-5 structural drawing.
The sample set of Fig. 3 convolutional neural networks training.
Embodiment
The present invention is further described below in conjunction with accompanying drawing.Be illustrated in figure 1 as the process flow diagram based on the vehicle front trafficability analytical approach of convolutional neural networks.The present invention is example with the structured environment of highway, and environment before the car is divided into vehicle, road boundary and road surface.
Analytic process of the present invention comprises: the training of image acquisition, image pre-service, convolutional neural networks.
A, image acquisition
Camera acquisition by being installed in vehicle front a large amount of real highway running environment images (640 * 480 pixel), then with 3/5ths parts under the image as area-of-interest (640 * 288 pixel), to reduce follow-up workload; At last images cut is converted into gray level image.
B, image pre-service
The first step, the Gamma antidote has certain advantage aspect the minimizing illumination effect.Generally speaking, when the Gamma value greater than 1 the time, the high light part of image is compressed and shadow partly is expanded; When the Gamma value less than 1 the time, the high light of image partly is expanded and the shadow part is compressed.Utilize Gamma of method construct of nonlinear function stack to correct function, the gray level image that steps A obtains is corrected, functional expression is as follows:
G(x)＝1+f
1(x)+f
2(x)+f
3(x) （1）
f
1(x)＝acos(πx/255) （2）
f
2(x)＝(K(x)+b)cosβ+xsinα （3）
K(x)＝ρsin(4πx/255) （4）
α＝arctan(-2b/255) （5）
f
3(x)＝R(x)cos(3πx/255) （6）
R(x)＝c|2x/255-1| （7）
In the formula, x is a certain gray values of pixel points, the Gamma correction value of a certain gray-scale value correspondence of G (x) representative, and a ∈ (0,1) is a weighting coefficient, b represents f
2(x) maximum changing range, ρ are represented the amplitude of K (x), and α represents the deflection angle of K (x), and c represents the amplitude of R (x), and satisfy a+b+c＜1.Generally speaking, a＜b≤c, desirable a=0.2, b=0.3, c=0.3.
Gray-scale value computing formula after Gamma corrects is:
g(x)＝255(x/255)
1/G(x) （8）
In the formula, a certain gray values of pixel points after g (x) representative is corrected through Gamma.
Correct through Gamma, obtain gray level image P.
Second step, at the gray level image P that corrects through Gamma, choose the gray-scale value except vehicle and road boundary in the image and be 0 pixel, change these gray values of pixel points into 1 by programming, choose the gray-scale value except vehicle and road boundary in the image and be 255 pixel, change these gray values of pixel points into 254 by programming; Change vehicle region gray values of pixel points in the image into 0 by programming then, the gray-scale value of road boundary area pixel point changes 255 into, obtains gray level image Q.So far, the pixel of gray level image Q comprises three classes: the first kind is that gray-scale value is 0 pixel, represents vehicle; Second class is that gray-scale value is 255 pixel, represents road boundary; The 3rd class is that to remove gray-scale value be pixel outside 0 and 255, represents the road surface.Give corresponding label respectively by being programmed for above three class pixels, be about to label " 0 " and compose expression " vehicle " to first kind pixel, label " 1 " is composed to the second class pixel, expression " road boundary " composes label " 2 " to the 3rd class pixel, expression " road surface ".At last the label of each pixel among the image Q is composed the respective pixel point of giving among the gray level image P by programming.
In the 3rd step, carry out normalized at the size of gray level image P:
In image, the shared number of pixels of same target is subjected to himself very big apart from the influence of video camera distance, and namely the shared number of pixels of arbitrary target is inversely proportional to the distance of itself and video camera.The present invention proposes a kind of new geometrical normalization method, carries out normalized at the size of gray level image, avoids reducing the resolution of image, reduces the differences in resolution that identification target range video camera distance causes.At first, on picture altitude, choose different pixel columns at interval, represent with vector x, x={15 32 60 65 68 72 75 82 85 87 92 96 100 108 111 113 124 130 138 143 150 160}, measure by actual samples, obtain the pixel wide of the corresponding target of different pixels row institute and highly; Secondly, be that 0～32 image-region is reference with pixels tall in the image, be benchmark with pixel wide W and the height H of target in this reference picture zone, namely the horizontal and vertical cutting scale-up factor in this reference picture zone is 1; At last, remove width and the height of target on all the other each pixel columns respectively with W and H, obtain two groups of ratios, represent with Y and Z respectively, by programming pixel column x data and Y and Z data are carried out match respectively then, obtain two matched curves, as follows:
Y＝0.0312x-0.8339 （9）
Z＝0.0360x-1.0590 （10）
Wherein, the horizontal cutting scale-up factor of Y representative image, the vertical cutting scale-up factor of Z representative image, a certain pixel column of x representative image.
The horizontal and vertical cutting ratio in reference picture zone is 1, is about to the image pattern that the reference picture zone is cut to 32 * 32 pixels.Along with the increase of x, according to the also corresponding increase of horizontal and vertical cutting size of formula (9), (10) acquisition.By cutting, obtain a series of image patterns not of uniform size, the image pattern that cutting is obtained is by the unified image that is scaled 32 * 32 pixels of programming at last.With the image of 32 * 32 pixels that the obtain training sample as convolutional neural networks.
The training of C, convolutional neural networks
The structure that the present invention adopts typical convolutional neural networks LeNet-5 as shown in Figure 2.Typical case's convolutional neural networks LeNet-5 forms by 8 layers, and input picture is 32 * 32 pixels; Network layer C1, C3, C5 represent convolutional layer, and network layer S2, S4 are time sampling layer, and network layer F5 is full articulamentum, and the neuronic number of output layer is identical with the target classification number that will identify, can change according to actual application environment.A face of every layer network represents a characteristic pattern, the set that the neuron that this characteristic pattern is shared by weights in each layer is formed.The neuron of each layer only is connected with the neuron of a local receptive field of last layer (counting from input layer).
Convolutional layer C1 is that 28 * 28 characteristic pattern is formed by 6 sizes, and each neuron of characteristic pattern is connected with one 5 * 5 neighborhood of input picture, can train and is connected with 122304 but convolutional layer C1 comprises 156 training parameters.Inferior sampling layer S2 is that 14 * 14 characteristic pattern is formed by 6 sizes, and size is that 2 * 2 neighborhood links to each other among each neuron of characteristic pattern and the convolutional layer C1, but an inferior sampling layer S2 has 12 training parameters can train and be connected with 5880.Convolutional layer C3 is that 10 * 10 characteristic pattern is formed by 16 sizes, and each neuron of characteristic pattern links to each other with time one 5 * 5 the neighborhood of sampling layer S2, can train and is connected with 151600 but convolutional layer C3 includes 1516 training parameters.Inferior sampling layer S4 is that 5 * 5 characteristic pattern is formed by 16 sizes, and each neuron of characteristic pattern and the size of convolutional layer C3 are that 2 * 2 neighborhood is connected, and can train and is connected with 2000 but an inferior sampling layer S4 includes 32 training parameters.Convolutional layer C5 is made up of 120 characteristic patterns, and each neuron of characteristic pattern is connected with time neighborhood of 5 * 5 sizes of all characteristic patterns of sampling layer S4, can train and is connected with 48120 but convolutional layer C5 includes 48120 training parameters.Network layer F6 is connected entirely with convolutional layer C5, but full articulamentum F6 comprises 10164 training parameters.Output layer is made up of the radial basis function unit.
The general type of convolutional layer is:
In the formula, l represents the number of plies, and k is convolution kernel, M
jRepresent the selection of input feature vector figure, b represents biasing.
The general type of inferior sampling layer is:
In the formula, down () expression time sampling function generally is the zone summation to a n * n of last tomographic image, and β represents time weights of sampling layer, and b represents biasing.
The present invention is example with the highway environment, environment is divided into road surface, vehicle and road boundary three parts before the car, therefore the output neuron number of LeNe-5 should be decided to be 3, and output 0 representative identification target is vehicle, output 1 representative identification target is road boundary, and output 2 representative identification targets are the road surface.The sample set size of network training is 5000, and the part sample as shown in Figure 3.Choosing of network initial weight adopts random approach to produce.With this known classification mode the convolution network is trained, network has just possessed the mapping ability between the inputoutput pair.By training, if satisfy the output valve of convolutional neural networks and the error of expectation value is in the tolerance interval, just obtained can be used for the convolutional neural networks of vehicle front trafficability analysis.
Claims (1)
1. vehicle front trafficability analytical approach based on convolutional neural networks is characterized in that: may further comprise the steps:
A, image acquisition
At first by a large amount of real vehicle running environment images of the camera acquisition that is installed in vehicle front, described image has m * n pixel; Obtain the zone of image bottom 3/5ths as area-of-interest by cutting then; At last images cut is converted into gray level image;
B, image pre-service
B1, the Gamma of method construct that utilizes nonlinear function to superpose correct function, and the gray level image that steps A obtains is corrected, and concrete functional expression is as follows:
G(x)＝1+f
1(x)+f
2(x)+f
3(x) （1）
f
1(x)＝acos(πx/255) （2）
f
2(x)＝(K(x)+b)cosβ+xsinα （3）
K(x)＝ρsin(4πx/255) （4）
α＝arctan(-2b/255) （5）
f
3(x)＝R(x)cos(3πx/255) （6）
R(x)＝c|2x/255-1| （7）
In the formula, x is a certain gray values of pixel points, the Gamma correction value of a certain gray-scale value correspondence of G (x) representative, and a ∈ (0,1) is a weighting coefficient, b represents f
2(x) maximum changing range, ρ are represented the amplitude of K (x), and α represents the deflection angle of K (x), and c represents the amplitude of R (x), and satisfy a+b+c＜1;
Gray-scale value computing formula after Gamma corrects is:
g(x)＝255(x/255)
1/G(x) （8）
In the formula, a certain gray values of pixel points after g (x) representative is corrected through Gamma;
Correct through Gamma, obtain gray level image P;
B2, B2, at gray level image P, change some gray values of pixel points, concrete change method is as follows:
Choose in the image gray-scale value in the image-region except vehicle and road boundary and be 0 pixel, change its gray-scale value into 1, choose in the image gray-scale value in the image-region except vehicle and road boundary and be 255 pixel, change its gray-scale value into 254; Change vehicle region gray values of pixel points in the image into 0, the gray-scale value of road boundary area pixel point changes 255 into, and the image behind the change pixel is gray level image Q; So far, the pixel of gray level image Q comprises three classes: the first kind is that gray-scale value is 0 pixel, represents vehicle; Second class is that gray-scale value is 255 pixel, represents road boundary; The 3rd class is that to remove gray-scale value be pixel outside 0 and 255, represents the road surface; Give corresponding label respectively with above three class pixels, be about to label " 0 " and compose expression " vehicle " to first kind pixel, label " 1 " is composed to the second class pixel, expression " road boundary " composes label " 2 " to the 3rd class pixel, expression " road surface "; At last the label of each pixel among the gray level image Q is composed to corresponding pixel among the gray level image P;
B3, carry out normalized at the size of gray level image P:
B31, along the picture altitude direction, at interval choose different pixel columns, represent with x, measure by actual samples, obtain the pixel wide of the corresponding target of different pixels row x institute and highly;
B32, being that 0～32 image-region is the reference picture zone with pixels tall in the image, is benchmark with pixel wide W and the height H of the target that will identify in this reference picture zone, and the horizontal and vertical cutting scale-up factor of namely establishing this reference picture zone is 1; With W and H remove respectively target on all the other each pixel columns width and the height, obtain two groups of ratios, represent with Y and Z respectively;
B33, at last pixel column x and two groups of ratio Y and Z are carried out match respectively, obtain two matched curves, as follows:
Y＝k
1x+b
1 （9）
Z＝k
2x+b
2 （10）
Wherein, the horizontal cutting scale-up factor of Y representative image, the vertical cutting scale-up factor of Z representative image, a certain pixel column of x representative image, k
1, k
2Represent the slope of two matched curves respectively, b
1, b
2Represent the intercept of two matched curves respectively;
B34, the horizontal and vertical cutting ratio in reference picture zone all is made as 1, is about to the image pattern that the reference picture zone is cut to 32 * 32 pixels; Along with the increase of x, according to the also corresponding increase of horizontal and vertical cutting size of formula (9), (10) acquisition; By cutting, obtain a series of image patterns not of uniform size, the image pattern unification that at last cutting is obtained is scaled the image of 32 * 32 pixels; With the image of 32 * 32 pixels that the obtain training sample as convolutional neural networks;
The training of C, convolutional neural networks
Typical case's convolutional neural networks LeNet-5 forms by 8 layers, and input is that layer is 32 * 32 pixel images; Network layer C1, C3 and C5 represent convolutional layer respectively, and network layer S2 and S4 are time sampling layer, and network layer F5 is full articulamentum, and the neuronic number of output layer is identical with the target classification number that will identify, changes according to actual application environment; A face of every layer network represents a characteristic pattern, the set that the neuron that this characteristic pattern is shared by weights in each layer is formed; The neuron of each layer only is connected with the neuron of a local receptive field of last layer;
The general type of convolutional layer is:
In the formula, { 1,2,3,4,5,6,7,8} represents the number of plies to l ∈, and k is convolution kernel, M
jRepresent the selection of input feature vector figure, b represents biasing;
The general type of inferior sampling layer is:
In the formula, down () expression time sampling function generally is the zone summation to a n * n of last tomographic image, and β represents time weights of sampling layer, and b represents biasing;
According to actual application environment, the number of the output neuron of LeNet-5 to be adjusted, the Pixel Dimensions that adopts step B to obtain then is that 32 * 32 image pattern is trained; By training, when the output valve of convolutional neural networks and the error of expectation value are in the tolerance interval, just obtain can be used for the convolutional neural networks of vehicle front trafficability analysis.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310234126.0A CN103279759B (en) | 2013-06-09 | 2013-06-09 | A kind of vehicle front trafficability analytical procedure based on convolutional neural networks |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201310234126.0A CN103279759B (en) | 2013-06-09 | 2013-06-09 | A kind of vehicle front trafficability analytical procedure based on convolutional neural networks |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN103279759A true CN103279759A (en) | 2013-09-04 |
| CN103279759B CN103279759B (en) | 2016-06-01 |
Family
ID=49062274
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201310234126.0A Expired - Fee Related CN103279759B (en) | 2013-06-09 | 2013-06-09 | A kind of vehicle front trafficability analytical procedure based on convolutional neural networks |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN103279759B (en) |
Cited By (34)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544506A (en) * | 2013-10-12 | 2014-01-29 | Tcl集团股份有限公司 | Method and device for classifying images on basis of convolutional neural network |
| CN104063719A (en) * | 2014-06-27 | 2014-09-24 | 深圳市赛为智能股份有限公司 | Method and device for pedestrian detection based on depth convolutional network |
| CN104077577A (en) * | 2014-07-03 | 2014-10-01 | 浙江大学 | Trademark detection method based on convolutional neural network |
| CN104182756A (en) * | 2014-09-05 | 2014-12-03 | 大连理工大学 | Method for detecting barriers in front of vehicles on basis of monocular vision |
| CN104680508A (en) * | 2013-11-29 | 2015-06-03 | 华为技术有限公司 | Convolutional neural network and target object detection method based on convolutional neural network |
| CN104809426A (en) * | 2014-01-27 | 2015-07-29 | 日本电气株式会社 | Convolutional neural network training method and target identification method and device |
| CN104850864A (en) * | 2015-06-01 | 2015-08-19 | 深圳英智源智能系统有限公司 | Unsupervised image recognition method based on convolutional neural network |
| CN104992179A (en) * | 2015-06-23 | 2015-10-21 | 浙江大学 | Fine-grained convolutional neural network-based clothes recommendation method |
| CN105224963A (en) * | 2014-06-04 | 2016-01-06 | 华为技术有限公司 | The method of changeable degree of depth learning network structure and terminal |
| CN105320495A (en) * | 2014-07-22 | 2016-02-10 | 英特尔公司 | Weight-shifting mechanism for convolutional neural network |
| CN105691381A (en) * | 2016-03-10 | 2016-06-22 | 大连理工大学 | Stability control method and system for electric automobile with four independently driven wheels |
| WO2016095117A1 (en) * | 2014-12-17 | 2016-06-23 | Nokia Technologies Oy | Object detection with neural network |
| CN105975915A (en) * | 2016-04-28 | 2016-09-28 | 大连理工大学 | Front vehicle parameter identification method based on multitask convolution nerve network |
| WO2016149881A1 (en) * | 2015-03-20 | 2016-09-29 | Intel Corporation | Object recogntion based on boosting binary convolutional neural network features |
| CN106205126A (en) * | 2016-08-12 | 2016-12-07 | 北京航空航天大学 | Large-scale Traffic Network based on convolutional neural networks is blocked up Forecasting Methodology and device |
| CN106407931A (en) * | 2016-09-19 | 2017-02-15 | 杭州电子科技大学 | Novel deep convolution neural network moving vehicle detection method |
| CN106599832A (en) * | 2016-12-09 | 2017-04-26 | 重庆邮电大学 | Method for detecting and recognizing various types of obstacles based on convolution neural network |
| CN106874296A (en) * | 2015-12-14 | 2017-06-20 | 阿里巴巴集团控股有限公司 | A kind of style recognition methods of commodity and device |
| CN107392189A (en) * | 2017-09-05 | 2017-11-24 | 百度在线网络技术（北京）有限公司 | For the method and apparatus for the driving behavior for determining unmanned vehicle |
| WO2017206066A1 (en) * | 2016-05-31 | 2017-12-07 | Nokia Technologies Oy | Method and apparatus for detecting small objects with an enhanced deep neural network |
| CN108122234A (en) * | 2016-11-29 | 2018-06-05 | 北京市商汤科技开发有限公司 | Convolutional neural networks training and method for processing video frequency, device and electronic equipment |
| CN108496178A (en) * | 2016-01-05 | 2018-09-04 | 御眼视觉技术有限公司 | System and method for estimating Future Path |
| CN108596115A (en) * | 2018-04-27 | 2018-09-28 | 济南浪潮高新科技投资发展有限公司 | A kind of vehicle checking method, apparatus and system based on convolutional neural networks |
| TWI638332B (en) * | 2016-11-29 | 2018-10-11 | 財團法人車輛研究測試中心 | Hierarchical object detection system with parallel architecture and method thereof |
| CN108921003A (en) * | 2018-04-26 | 2018-11-30 | 东华大学 | Unmanned plane obstacle detection method based on convolutional neural networks and morphological image |
| WO2019001346A1 (en) * | 2017-06-26 | 2019-01-03 | Huawei Technologies Co., Ltd. | System and methods for object filtering and uniform representation for autonomous systems |
| WO2019047649A1 (en) * | 2017-09-05 | 2019-03-14 | 百度在线网络技术（北京）有限公司 | Method and device for determining driving behavior of unmanned vehicle |
| CN109657643A (en) * | 2018-12-29 | 2019-04-19 | 百度在线网络技术（北京）有限公司 | A kind of image processing method and device |
| CN109726615A (en) * | 2017-10-30 | 2019-05-07 | 北京京东尚科信息技术有限公司 | A kind of recognition methods of road boundary and device |
| CN109961509A (en) * | 2019-03-01 | 2019-07-02 | 北京三快在线科技有限公司 | Dimensionally map generalization and model training method, device and electronic equipment |
| CN110084190A (en) * | 2019-04-25 | 2019-08-02 | 南开大学 | Unstructured road detection method in real time under a kind of violent light environment based on ANN |
| CN110399929A (en) * | 2017-11-01 | 2019-11-01 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN111222476A (en) * | 2020-01-10 | 2020-06-02 | 北京百度网讯科技有限公司 | Video time sequence action detection method and device, electronic equipment and storage medium |
| CN113325948A (en) * | 2020-02-28 | 2021-08-31 | 华为技术有限公司 | Air-isolated gesture adjusting method and terminal |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007027452A1 (en) * | 2005-08-31 | 2007-03-08 | Microsoft Corporation | Training convolutional neural networks on graphics processing units |
| CN101742341A (en) * | 2010-01-14 | 2010-06-16 | 中山大学 | Method and device for image processing |
| CN102750544A (en) * | 2012-06-01 | 2012-10-24 | 浙江捷尚视觉科技有限公司 | Detection system and detection method of rule-breaking driving that safety belt is not fastened and based on plate number recognition |
-
2013
- 2013-06-09 CN CN201310234126.0A patent/CN103279759B/en not_active Expired - Fee Related
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| WO2007027452A1 (en) * | 2005-08-31 | 2007-03-08 | Microsoft Corporation | Training convolutional neural networks on graphics processing units |
| CN101742341A (en) * | 2010-01-14 | 2010-06-16 | 中山大学 | Method and device for image processing |
| CN102750544A (en) * | 2012-06-01 | 2012-10-24 | 浙江捷尚视觉科技有限公司 | Detection system and detection method of rule-breaking driving that safety belt is not fastened and based on plate number recognition |
Non-Patent Citations (1)
| Title |
|---|
| 彭国福: "图像处理中Gamma校正的研究和实现", 《电子工程师》 * |
Cited By (52)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN103544506A (en) * | 2013-10-12 | 2014-01-29 | Tcl集团股份有限公司 | Method and device for classifying images on basis of convolutional neural network |
| CN103544506B (en) * | 2013-10-12 | 2017-08-08 | Tcl集团股份有限公司 | A kind of image classification method and device based on convolutional neural networks |
| CN104680508A (en) * | 2013-11-29 | 2015-06-03 | 华为技术有限公司 | Convolutional neural network and target object detection method based on convolutional neural network |
| CN104809426A (en) * | 2014-01-27 | 2015-07-29 | 日本电气株式会社 | Convolutional neural network training method and target identification method and device |
| CN104809426B (en) * | 2014-01-27 | 2019-04-05 | 日本电气株式会社 | Training method, target identification method and the device of convolutional neural networks |
| CN105224963B (en) * | 2014-06-04 | 2019-06-07 | 华为技术有限公司 | The method and terminal of changeable deep learning network structure |
| CN105224963A (en) * | 2014-06-04 | 2016-01-06 | 华为技术有限公司 | The method of changeable degree of depth learning network structure and terminal |
| CN104063719A (en) * | 2014-06-27 | 2014-09-24 | 深圳市赛为智能股份有限公司 | Method and device for pedestrian detection based on depth convolutional network |
| CN104077577A (en) * | 2014-07-03 | 2014-10-01 | 浙江大学 | Trademark detection method based on convolutional neural network |
| TWI635446B (en) * | 2014-07-22 | 2018-09-11 | 英特爾股份有限公司 | Weight-shifting appratus, method, system and machine accessible storage medium |
| CN105320495A (en) * | 2014-07-22 | 2016-02-10 | 英特尔公司 | Weight-shifting mechanism for convolutional neural network |
| CN104182756A (en) * | 2014-09-05 | 2014-12-03 | 大连理工大学 | Method for detecting barriers in front of vehicles on basis of monocular vision |
| CN104182756B (en) * | 2014-09-05 | 2017-04-12 | 大连理工大学 | Method for detecting barriers in front of vehicles on basis of monocular vision |
| US10275688B2 (en) | 2014-12-17 | 2019-04-30 | Nokia Technologies Oy | Object detection with neural network |
| WO2016095117A1 (en) * | 2014-12-17 | 2016-06-23 | Nokia Technologies Oy | Object detection with neural network |
| CN107004138A (en) * | 2014-12-17 | 2017-08-01 | 诺基亚技术有限公司 | Utilize the object detection of neutral net |
| WO2016149881A1 (en) * | 2015-03-20 | 2016-09-29 | Intel Corporation | Object recogntion based on boosting binary convolutional neural network features |
| US10685262B2 (en) | 2015-03-20 | 2020-06-16 | Intel Corporation | Object recognition based on boosting binary convolutional neural network features |
| CN104850864A (en) * | 2015-06-01 | 2015-08-19 | 深圳英智源智能系统有限公司 | Unsupervised image recognition method based on convolutional neural network |
| CN104992179A (en) * | 2015-06-23 | 2015-10-21 | 浙江大学 | Fine-grained convolutional neural network-based clothes recommendation method |
| CN106874296A (en) * | 2015-12-14 | 2017-06-20 | 阿里巴巴集团控股有限公司 | A kind of style recognition methods of commodity and device |
| CN108496178B (en) * | 2016-01-05 | 2023-08-08 | 御眼视觉技术有限公司 | System and method for estimating future path |
| CN108496178A (en) * | 2016-01-05 | 2018-09-04 | 御眼视觉技术有限公司 | System and method for estimating Future Path |
| CN105691381A (en) * | 2016-03-10 | 2016-06-22 | 大连理工大学 | Stability control method and system for electric automobile with four independently driven wheels |
| CN105691381B (en) * | 2016-03-10 | 2018-04-27 | 大连理工大学 | A kind of four motorized wheels electric automobile stability control method and system |
| CN105975915A (en) * | 2016-04-28 | 2016-09-28 | 大连理工大学 | Front vehicle parameter identification method based on multitask convolution nerve network |
| CN105975915B (en) * | 2016-04-28 | 2019-05-21 | 大连理工大学 | A kind of front vehicles parameter identification method based on multitask convolutional neural networks |
| US11631005B2 (en) | 2016-05-31 | 2023-04-18 | Nokia Technologies Oy | Method and apparatus for detecting small objects with an enhanced deep neural network |
| WO2017206066A1 (en) * | 2016-05-31 | 2017-12-07 | Nokia Technologies Oy | Method and apparatus for detecting small objects with an enhanced deep neural network |
| CN106205126A (en) * | 2016-08-12 | 2016-12-07 | 北京航空航天大学 | Large-scale Traffic Network based on convolutional neural networks is blocked up Forecasting Methodology and device |
| CN106407931A (en) * | 2016-09-19 | 2017-02-15 | 杭州电子科技大学 | Novel deep convolution neural network moving vehicle detection method |
| CN106407931B (en) * | 2016-09-19 | 2019-11-22 | 杭州电子科技大学 | A kind of depth convolutional neural networks moving vehicle detection method |
| TWI638332B (en) * | 2016-11-29 | 2018-10-11 | 財團法人車輛研究測試中心 | Hierarchical object detection system with parallel architecture and method thereof |
| CN108122234A (en) * | 2016-11-29 | 2018-06-05 | 北京市商汤科技开发有限公司 | Convolutional neural networks training and method for processing video frequency, device and electronic equipment |
| CN106599832A (en) * | 2016-12-09 | 2017-04-26 | 重庆邮电大学 | Method for detecting and recognizing various types of obstacles based on convolution neural network |
| WO2019001346A1 (en) * | 2017-06-26 | 2019-01-03 | Huawei Technologies Co., Ltd. | System and methods for object filtering and uniform representation for autonomous systems |
| WO2019047649A1 (en) * | 2017-09-05 | 2019-03-14 | 百度在线网络技术（北京）有限公司 | Method and device for determining driving behavior of unmanned vehicle |
| WO2019047655A1 (en) * | 2017-09-05 | 2019-03-14 | 百度在线网络技术（北京）有限公司 | Method and apparatus for use in determining driving behavior of driverless vehicle |
| CN107392189A (en) * | 2017-09-05 | 2017-11-24 | 百度在线网络技术（北京）有限公司 | For the method and apparatus for the driving behavior for determining unmanned vehicle |
| CN109726615A (en) * | 2017-10-30 | 2019-05-07 | 北京京东尚科信息技术有限公司 | A kind of recognition methods of road boundary and device |
| CN110399929A (en) * | 2017-11-01 | 2019-11-01 | 腾讯科技（深圳）有限公司 | Eye fundus image classification method, device and computer readable storage medium |
| CN110399929B (en) * | 2017-11-01 | 2023-04-28 | 腾讯科技（深圳）有限公司 | Fundus image classification method, fundus image classification apparatus, and computer-readable storage medium |
| CN108921003A (en) * | 2018-04-26 | 2018-11-30 | 东华大学 | Unmanned plane obstacle detection method based on convolutional neural networks and morphological image |
| CN108596115A (en) * | 2018-04-27 | 2018-09-28 | 济南浪潮高新科技投资发展有限公司 | A kind of vehicle checking method, apparatus and system based on convolutional neural networks |
| CN109657643A (en) * | 2018-12-29 | 2019-04-19 | 百度在线网络技术（北京）有限公司 | A kind of image processing method and device |
| CN109961509A (en) * | 2019-03-01 | 2019-07-02 | 北京三快在线科技有限公司 | Dimensionally map generalization and model training method, device and electronic equipment |
| CN110084190A (en) * | 2019-04-25 | 2019-08-02 | 南开大学 | Unstructured road detection method in real time under a kind of violent light environment based on ANN |
| CN110084190B (en) * | 2019-04-25 | 2024-02-06 | 南开大学 | Real-time unstructured road detection method under severe illumination environment based on ANN |
| CN111222476A (en) * | 2020-01-10 | 2020-06-02 | 北京百度网讯科技有限公司 | Video time sequence action detection method and device, electronic equipment and storage medium |
| US11600069B2 (en) | 2020-01-10 | 2023-03-07 | Beijing Baidu Netcom Science And Technology Co., Ltd. | Method and apparatus for detecting temporal action of video, electronic device and storage medium |
| CN113325948B (en) * | 2020-02-28 | 2023-02-07 | 华为技术有限公司 | Air-isolated gesture adjusting method and terminal |
| CN113325948A (en) * | 2020-02-28 | 2021-08-31 | 华为技术有限公司 | Air-isolated gesture adjusting method and terminal |
Also Published As
| Publication number | Publication date |
|---|---|
| CN103279759B (en) | 2016-06-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN103279759A (en) | Vehicle front trafficability analyzing method based on convolution nerve network | |
| CN103034863B (en) | The remote sensing image road acquisition methods of a kind of syncaryon Fisher and multiple dimensioned extraction | |
| CN107316064B (en) | Asphalt pavement crack classification and identification method based on convolutional neural network | |
| CN103093250B (en) | A kind of Adaboost method for detecting human face based on new Haar-like feature | |
| CN104182985B (en) | Remote sensing image change detection method | |
| CN103020605B (en) | Bridge identification method based on decision-making layer fusion | |
| CN103729853B (en) | High score remote sensing image building under three-dimension GIS auxiliary damages detection method | |
| CN105184271A (en) | Automatic vehicle detection method based on deep learning | |
| CN102968790B (en) | Remote sensing image change detection method based on image fusion | |
| CN103902976A (en) | Pedestrian detection method based on infrared image | |
| CN103413151A (en) | Hyperspectral image classification method based on image regular low-rank expression dimensionality reduction | |
| CN103679205B (en) | Assume based on shade and the Foregut fermenters method of layering HOG symmetrical feature checking | |
| CN104299006A (en) | Vehicle license plate recognition method based on deep neural network | |
| CN103218831A (en) | Video moving target classification and identification method based on outline constraint | |
| CN104867150A (en) | Wave band correction change detection method of remote sensing image fuzzy clustering and system thereof | |
| CN103632363A (en) | Object-level high-resolution remote sensing image change detection method based on multi-scale fusion | |
| CN104361351A (en) | Synthetic aperture radar (SAR) image classification method on basis of range statistics similarity | |
| CN105005989A (en) | Vehicle target segmentation method under weak contrast | |
| CN102867183B (en) | Method and device for detecting littered objects of vehicle and intelligent traffic monitoring system | |
| CN102629380B (en) | Remote sensing image change detection method based on multi-group filtering and dimension reduction | |
| CN104751166A (en) | Spectral angle and Euclidean distance based remote-sensing image classification method | |
| Yuan et al. | Learning to count buildings in diverse aerial scenes | |
| CN106096655A (en) | A kind of remote sensing image airplane detection method based on convolutional neural networks | |
| CN104103070B (en) | Landing point selecting method based on optical images | |
| CN103488993A (en) | Crowd abnormal behavior identification method based on FAST |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| C06 | Publication | ||
| PB01 | Publication | ||
| C10 | Entry into substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| C14 | Grant of patent or utility model | ||
| GR01 | Patent grant | ||
| CF01 | Termination of patent right due to non-payment of annual fee |
Granted publication date: 20160601 Termination date: 20190609 |
|
| CF01 | Termination of patent right due to non-payment of annual fee |