CN106874945A - A kind of pavement traffic lights detecting system and method for visually impaired people - Google Patents
A kind of pavement traffic lights detecting system and method for visually impaired people Download PDFInfo
- Publication number
- CN106874945A CN106874945A CN201710056650.1A CN201710056650A CN106874945A CN 106874945 A CN106874945 A CN 106874945A CN 201710056650 A CN201710056650 A CN 201710056650A CN 106874945 A CN106874945 A CN 106874945A
- Authority
- CN
- China
- Prior art keywords
- area
- traffic lights
- detection zone
- rectangular area
- 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.)
- Granted
Links
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
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/40—Extraction of image or video features
- G06V10/56—Extraction of image or video features relating to colour
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)
- Multimedia (AREA)
- Image Analysis (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention discloses a kind of pavement traffic lights detection method for visually impaired people.The method gathers image using camera, and the image for gathering is processed using compact processor, exports the state and actual range of pavement traffic lights.The method can detect the pavement traffic lights under the different illumination conditions such as daytime, night, and the method false drop rate is low, loss is low, real-time is good, professional platform independence is good.The requirement that visually impaired people goes across the road can well be met.
Description
Technical field
The invention belongs to mode identification technology, image processing techniques, visually impaired people's ancillary technique field, it is related to a kind of people's row
Road traffic lights detecting system and method.
Background technology
Visual information be the mankind identification surrounding environment most important information source, the mankind obtain information 80% or so be from
Vision system input.Counted according to the World Health Organization, the whole world there are 2.85 hundred million dysopia personages.Visually impaired people have lost
Normal vision, the understanding to color, shape is highly difficult.Now, many people in them are assisted using empty-handed cane or seeing-eye dog
The daily life of oneself.Obviously, empty-handed cane is not enough to solve all of difficulty during travelling.Seeing-eye dog can guide visually impaired people
Scholar is to avoid the danger during walking on road, but because the very big cost of training seeing-eye dog needs, they cannot be used for institute
There is visually impaired person.Therefore, the conventional tool such as walking stick, seeing-eye dog cannot offer is sufficient to assist for they go on a journey.
Since electronics trip auxiliary (ETA) equipment development, a kind of auxiliary visually impaired person has been considered as it in varied situations
The effective method of trip.In order to help user to find path, many accessory system deployment depth cameras come detect can and road
Footpath and obstacle.However, the identification of specific project, such as traffic lights are detected, almost it is not integrated in all these systems.Due to
Eyesight is limited, and visually impaired person is felt to be difficult to pass through road.There are many pedestrian's traffic lights in urban district, but without all for visually impaired people matches somebody with somebody
Standby sound auxiliary equipment.Therefore, for visually impaired person avoids road hazard, detection pedestrian traffic lamp is most important.
Many work are devoted to solving traffic lights test problems.However, most of current solutions are applied to independently
Automobile navigation.In these cases, camera is static relative to bearer.Blind person's assistance application is completely different, crossing
With any position that traffic lights may be located at image, and by handheld camera capture video be unstable.Vehicular traffic lamp
With circular or arrowhead form, background is also very simple, such as sky.But pavement traffic lights has complicated shape, Er Qie
It is not significant in background.Therefore, the pavement traffic lights detection algorithm for visually impaired person must be sufficiently robust, various to tackle
Environment.Used as the householder method to visually impaired people, detection algorithm should have low-down false alarm rate, this safety for user
It is highly important.In addition, real-time is also a requirement of algorithm.Because algorithm must be realized in portable stage, have
The system resource of limit needs effective algorithm to maintain appropriate frame rate.
The content of the invention
The purpose of the present invention is to solve the shortcomings of the prior art, there is provided a kind of pavement traffic lights detecting system and side
Method.
The purpose of the present invention is achieved through the following technical solutions：A kind of pavement traffic lights detecting system, the system
System includes a color camera, a compact processor, a handset module, a battery module.Color camera and small-sized place
Reason device is connected, and battery module is connected with compact processor.Color camera gathers the coloured image of surrounding scene, small-sized place in real time
Reason device is processed the coloured image for obtaining, and the traffic lights information that will be recognized is converted into voice signal, and be transmitted to earphone mould
Block, and then to user.
The pavement traffic lights detection method of said system is as follows:
(1) the coloured image Color that color camera is collected, similar to Fig. 3, is transmitted to compact processor and is processed.
(2) nonqualifying pixels extraction is carried out in HSV space.The span of HSV space triple channel be Hue ∈ [0,360),
Saturation ∈ [0,256), and Value ∈ [0,256), the nonqualifying pixels are to meet Value>Va also, and Hue<Hu1 or
Person Hue>The pixel of hu2；Wherein 50<va<256,250>hu1>230,250<hu2<300.
(3) to all nonqualifying pixels, connected region is extracted using region-growing method, each connected region is used as a time
Favored area, calculates candidate region area a；
(4) for any one candidate region, the minimum rectangular area for covering the candidate region, and the rectangular area are set up
Every one side parallel to coloured image side.And the height and width of the minimum rectangular area are extracted, obtain the minimum rectangular area
Depth-width ratio r, and filling rate f (candidate region and minimum rectangular area area ratio)；Further to candidate region by size a
(connected domain area), the screening of depth-width ratio r and filling rate f filters out a1<a<a2、r1<r<The qualified candidate of r2 and f ＞ f1
Region, wherein, 10-6area<a1<10-4Area, 10-3area<a2<10-1area,0<r1<1,1.5<r2<4,0.2<f1<0.6,
Wherein area is Color areas.
(5) it is extension center with the center of minimum rectangular area, minimum rectangular area is arrived wide grade of height than expanding to three
Six times, obtain rectangular extension region.In calculating the extension rectangular area, beyond the mean flow rate v1 of candidate region and candidate region
Background area mean flow rate v2, filter out v1>The rectangular extension region of v2 is used as detection zone.
(6) detection zone that obtains of step 5 screening is carried out into gray processing treatment, and unified pixel size, extract HOG or
LBP characteristics of image.
(7) according to characteristics of image, detection zone is predicted by SVM models, is predicted the outcome；The prediction knot
Fruit is selected from：A () red or yellow pavement traffic lights, (b) green pavement traffic lights, (c) non-pavement traffic lights.
(8) predicting the outcome and verified for step 7 output, predicts the outcome as in the case of (b), if detection zone color
Phase average h1<Hue<h2, then end product is green light；Otherwise it is judged to non-pavement traffic lights.Wherein parameter value scope is
100<h1<180,150<h2<220.Predict the outcome as in the case of (a), if detection zone parcel form and aspect average wh1<WH<wh2,
Then end product is red light.If detection zone parcel form and aspect average wh3<WH<wh4, then end product is amber light；Otherwise it is judged to
Non- pavement traffic lights.Wherein parameter value scope is 90<wh1<140,170<wh2<210；210<wh3<260,200<wh4<
290.Parcel form and aspect (WH), is defined as WH=(Hue+180) mod 360.
(9) expected areas are set up to each detection zone in the n-th-m two field pictures, expected areas set is constituted, and to every
It is m testing result sequences that individual expected areas build a capacity；Expected areas set is constantly updated by the following method,
Until present frame n：
All detection zones in each frame are matched with the expected areas set of former frame.If the center of detection zone
Then it is m using the newly-built expected areas of the detection zone and corresponding capacity not in the range of any one expected areas
Testing result sequence.If the center of detection zone is in the expected areas of former frame, with the expected areas of the detection zone more
The expected areas matched in new previous frame, and the testing result of the detection zone is added in testing result sequence.
The method for building up of the expected areas is：It is extension center with the center of the minimum rectangular area of the detection zone,
Width high etc. is carried out to minimum rectangular area than extending what is obtained after 30-60 times.
(10) testing result stored in the corresponding testing result sequence of each expected areas is counted, confidence level is calculatedWherein signal represents one of green three kinds of testing results of reddish yellow,
(11) that color of selection maximum confidence is exported as final traffic lights testing result, as shown in Figure 4.
(12) output result passes to blind person by handset module with the form of voice.
SVM training patterns in step 7, obtain by the following method：
(1) coloured image Color of the color camera collection comprising green pavement traffic lights, is transmitted at compact processor
Reason.
(2) nonqualifying pixels extraction is carried out in HSV space.The span of HSV space triple channel be Hue ∈ [0,360),
Saturation ∈ [0,256), and Value ∈ [0,256), the nonqualifying pixels are to meet Value>Va also, and Hue<Hu1 or
Person Hue>The pixel of hu2；Wherein 50<va<256,250>hu1>230,250<hu2<300.
(3) to all nonqualifying pixels, connected region is extracted using region-growing method, each connected region is used as a time
Favored area, calculates candidate region area a；
(4) for any one candidate region, the minimum rectangular area for covering the candidate region, and the rectangular area are set up
Every one side parallel to coloured image side.And the height and width of the minimum rectangular area are extracted, obtain the minimum rectangular area
Depth-width ratio r, and filling rate f (candidate region and minimum rectangular area area ratio)；Further to candidate region by size a
(connected domain area), the screening of depth-width ratio r and filling rate f filters out a1<a<a2、r1<r<The qualified candidate of r2 and f ＞ f1
Region, wherein, 10-6area<a1<10-4Area, 10-3area<a2<10-1area,0<r1<1,1.5<r2<4,0.2<f1<0.6,
Wherein area is Color areas.
(5) it is extension center with the center of minimum rectangular area, minimum rectangular area is arrived wide grade of height than expanding to three
Six times, obtain rectangular extension region.In calculating the extension rectangular area, beyond the mean flow rate v1 of candidate region and candidate region
Background area mean flow rate v2, filter out v1>The rectangular extension region of v2 is used as detection zone.
(6) detection zone that obtains of step 5 screening is carried out into gray processing treatment, and unified pixel size, extract HOG or
LBP characteristics of image.
(7) HOG the or LBP characteristics of image of yellow pavement traffic lights and rising star trade traffic lights is obtained according to step 1-6.
Data set to having marked is trained using SVM, and kernel is LINEAR or RBF functions, is obtained using cross validation method
To optimal precision and recall rate.
This method is essentially consisted in compared to the advantage of conventional pavement traffic lights detection method：
1st, error detection is low, and recall rate is high.This method can accurately obtain the pavement traffic lights in image, can accurately arrange
Except to interfering objects such as driveway traffic lights, car light, advertisement LED, and traffic lights can be detected from mixed and disorderly background.
2nd, ambient adaptability.This method can to high light, dim light, the varying environment such as night, white night look after under the conditions of people
Trade traffic lights is accurately detected.
3rd, real-time is good.This method (such as mobile phone) can in real time analyze pavement traffic lights on a mobile platform, each
Without postponing under the situation of kind, so as to ensure that visually impaired people's security.
4th, it is portable good.The system core part is camera, processor and earphone returning equipment, can be conveniently transplanted to
For on the smart machines such as mobile phone, flat board.
Brief description of the drawings
Fig. 1 is the module connection diagram of visually impaired people pavement traffic lights detecting system；
Fig. 2 is the structural representation of visually impaired people pavement traffic lights detecting system；
Fig. 3 is some original color image schematic diagrames
Fig. 4 is pavement traffic lights Detection results, and corresponding detection knot is represented with corresponding red, green or yellow circular color lump
Really.
Specific embodiment
As shown in figure 1, a kind of pavement traffic lights detecting system, the system includes a color camera, and one small-sized
Processor, a handset module, a battery module.Color camera is connected with compact processor, battery module and small-sized treatment
Device is connected.Color camera gathers the coloured image of surrounding scene in real time, and compact processor is at the coloured image that obtains
Reason, the traffic lights information that will be recognized is converted into voice signal, and is transmitted to handset module, and then to user.The system can set
Count into similar to the glasses described in Fig. 2, to reach aesthetic.
The pavement traffic lights detection method of the system is comprised the following steps:
(1) the coloured image Color that color camera is collected, similar to Fig. 3, is transmitted to compact processor and is processed.
(2) nonqualifying pixels extraction is carried out in HSV space.The span of HSV space triple channel be Hue ∈ [0,360),
Saturation ∈ [0,256), and Value ∈ [0,256), the nonqualifying pixels are to meet Value>Va also, and Hue<Hu1 or
Person Hue>The pixel of hu2；Wherein 50<va<256,250>hu1>230,250<hu2<300.
(3) to all nonqualifying pixels, connected region is extracted using region-growing method, each connected region is used as a time
Favored area, calculates candidate region area a；
(4) for any one candidate region, the minimum rectangular area for covering the candidate region, and the rectangular area are set up
Every one side parallel to coloured image side.And the height and width of the minimum rectangular area are extracted, obtain the minimum rectangular area
Depth-width ratio r, and filling rate f (candidate region and minimum rectangular area area ratio)；Further to candidate region by size a
(connected domain area), the screening of depth-width ratio r and filling rate f filters out a1<a<a2、r1<r<The qualified candidate of r2 and f ＞ f1
Region, wherein, 10-6area<a1<10-4Area, 10-3area<a2<10-1area,0<r1<1,1.5<r2<4,0.2<f1<0.6,
Wherein area is Color areas.
(5) it is extension center with the center of minimum rectangular area, minimum rectangular area is arrived wide grade of height than expanding to three
Six times, obtain rectangular extension region.In calculating the extension rectangular area, beyond the mean flow rate v1 of candidate region and candidate region
Background area mean flow rate v2, filter out v1>The rectangular extension region of v2 is used as detection zone.
(6) detection zone that obtains of step 5 screening is carried out into gray processing treatment, and unified pixel size, extract HOG or
LBP characteristics of image.
(7) according to characteristics of image, detection zone is predicted by SVM models, is predicted the outcome；The prediction knot
Fruit is selected from：A () red or yellow pavement traffic lights, (b) green pavement traffic lights, (c) non-pavement traffic lights.
(8) predicting the outcome and verified for step 7 output, predicts the outcome as in the case of (b), if detection zone color
Phase average h1<Hue<h2, then end product is green light；Otherwise it is judged to non-pavement traffic lights.Wherein parameter value scope is
100<h1<180,150<h2<220.Predict the outcome as in the case of (a), if detection zone parcel form and aspect average wh1<WH<wh2,
Then end product is red light.If detection zone parcel form and aspect average wh3<WH<wh4, then end product is amber light；Otherwise it is judged to
Non- pavement traffic lights.Wherein parameter value scope is 90<wh1<140,170<wh2<210；210<wh3<260,200<wh4<
290.Parcel form and aspect (WH), is defined as WH=(Hue+180) mod 360.
(9) expected areas are set up to each detection zone in the n-th-m two field pictures, expected areas set is constituted, and to every
It is m testing result sequences that individual expected areas build a capacity；Expected areas set is constantly updated by the following method,
Until present frame n：
All detection zones in each frame are matched with the expected areas set of former frame.If the center of detection zone
Then it is m using the newly-built expected areas of the detection zone and corresponding capacity not in the range of any one expected areas
Testing result sequence.If the center of detection zone is in the expected areas of former frame, with the expected areas of the detection zone more
The expected areas matched in new previous frame, and the testing result of the detection zone is added in testing result sequence.
The method for building up of the expected areas is：It is extension center with the center of the minimum rectangular area of the detection zone,
Width high etc. is carried out to minimum rectangular area than extending what is obtained after 30-60 times.
(10) testing result stored in the corresponding testing result sequence of each expected areas is counted, confidence level is calculatedWherein signal represents one of green three kinds of testing results of reddish yellow,
(11) that color of selection maximum confidence is exported as final traffic lights testing result, as shown in Figure 4.
(12) output result passes to blind person by handset module with the form of voice.
SVM training patterns in step 7, obtain by the following method：
(1) coloured image Color of the color camera collection comprising green pavement traffic lights, is transmitted at compact processor
Reason.
(2) nonqualifying pixels extraction is carried out in HSV space.The span of HSV space triple channel be Hue ∈ [0,360),
Saturation ∈ [0,256), and Value ∈ [0,256), the nonqualifying pixels are to meet Value>Va also, and Hue<Hu1 or
Person Hue>The pixel of hu2；Wherein 50<va<256,250>hu1>230,250<hu2<300.
(3) to all nonqualifying pixels, connected region is extracted using region-growing method, each connected region is used as a time
Favored area, calculates candidate region area a；
(4) for any one candidate region, the minimum rectangular area for covering the candidate region, and the rectangular area are set up
Every one side parallel to coloured image side.And the height and width of the minimum rectangular area are extracted, obtain the minimum rectangular area
Depth-width ratio r, and filling rate f (candidate region and minimum rectangular area area ratio)；Further to candidate region by size a
(connected domain area), the screening of depth-width ratio r and filling rate f filters out a1<a<a2、r1<r<The qualified candidate of r2 and f ＞ f1
Region, wherein, 10-6area<a1<10-4Area, 10-3area<a2<10-1area,0<r1<1,1.5<r2<4,0.2<f1<0.6,
Wherein area is Color areas.
(5) it is extension center with the center of minimum rectangular area, minimum rectangular area is arrived wide grade of height than expanding to three
Six times, obtain rectangular extension region.In calculating the extension rectangular area, beyond the mean flow rate v1 of candidate region and candidate region
Background area mean flow rate v2, filter out v1>The rectangular extension region of v2 is used as detection zone.
(6) detection zone that obtains of step 5 screening is carried out into gray processing treatment, and unified pixel size, extract HOG or
LBP characteristics of image.
(7) HOG the or LBP characteristics of image of yellow pavement traffic lights and rising star trade traffic lights is obtained according to step 1-6.
Claims (3)
1. a kind of pavement traffic lights detecting system for visually impaired people, the system includes a color camera, and one small
Type processor, a handset module, a battery module.Color camera is connected with compact processor, battery module and small-sized place
Reason device is connected.Color camera gathers the coloured image of surrounding scene in real time, and compact processor is carried out to the coloured image for obtaining
Treatment, the traffic lights information that will be recognized is converted into voice signal, and is transmitted to handset module, and then to user.
2. the pavement traffic lights detection method of system described in a kind of claim 1, it is characterised in that comprise the following steps：
(1) the coloured image Color that color camera is collected, is transmitted to compact processor and is processed.
(2) nonqualifying pixels extraction is carried out in HSV space.The span of HSV space triple channel be Hue ∈ [0,360),
Saturation ∈ [0,256), and Value ∈ [0,256), the nonqualifying pixels are to meet Value>Va also, and Hue<Hu1 or
Person Hue>The pixel of hu2；Wherein 50<va<256,250>hu1>230,250<hu2<300.
(3) to all nonqualifying pixels, connected region is extracted using region-growing method, each connected region is used as a candidate regions
Domain, calculates candidate region area a；
(4) for any one candidate region, the minimum rectangular area for covering the candidate region is set up, and the rectangular area is every
While parallel to the side of coloured image.And the height and width of the minimum rectangular area are extracted, the height for obtaining the minimum rectangular area is wide
Than r, and filling rate f (candidate region and minimum rectangular area area ratio)；Further candidate region (is connected by size a
Logical domain area), the screening of depth-width ratio r and filling rate f filters out a1<a<a2、r1<r<The qualified candidate regions of r2 and f ＞ f1
Domain, wherein, 10-6area<a1<10-4Area, 10-3area<a2<10-1area,0<r1<1,1.5<r2<4,0.2<f1<0.6, its
Middle area is Color areas.
(5) it is extension center with the center of minimum rectangular area, three to six times is expanded to by high grade ratio wide to minimum rectangular area,
Obtain rectangular extension region.In calculating the extension rectangular area, the back of the body beyond the mean flow rate v1 of candidate region and candidate region
The mean flow rate v2 of scene area, filters out v1>The rectangular extension region of v2 is used as detection zone.
(6) detection zone that step 5 screening is obtained is carried out into gray processing treatment, and unified pixel size, extracts HOG or LBP figures
As feature.
(7) according to characteristics of image, detection zone is predicted by SVM models, is predicted the outcome；The choosing that predicts the outcome
From：A () red or yellow pavement traffic lights, (b) green pavement traffic lights, (c) non-pavement traffic lights.
(8) predicting the outcome and verified for step 7 output, predicts the outcome as in the case of (b), if detection zone form and aspect are equal
Value h1<Hue<h2, then end product is green light；Otherwise it is judged to non-pavement traffic lights.Wherein parameter value scope is 100<h1
<180,150<h2<220.Predict the outcome as in the case of (a), if detection zone parcel form and aspect average wh1<WH<wh2, then finally
Result is red light.If detection zone parcel form and aspect average wh3<WH<wh4, then end product is amber light；Otherwise it is judged to inhuman row
Road traffic lights.Wherein parameter value scope is 90<wh1<140,170<wh2<210；210<wh3<260,200<wh4<290.Parcel
Form and aspect (WH), are defined as WH=(Hue+180) mod 360.
(9) expected areas are set up to each detection zone in the n-th-m two field pictures, expected areas set is constituted, and it is pre- to each
It is m testing result sequences that term area builds a capacity；Expected areas set is constantly updated by the following method, until
Present frame n：
All detection zones in each frame are matched with the expected areas set of former frame.If the center of detection zone does not exist
In the range of any one expected areas, then using the newly-built expected areas of the detection zone and corresponding capacity for m is detected
As a result sequence.If the center of detection zone is in the expected areas of former frame, updated with the expected areas of the detection zone
The expected areas matched in one frame, and the testing result of the detection zone is added in testing result sequence.
The method for building up of the expected areas is：It is extension center with the center of the minimum rectangular area of the detection zone, to most
Small rectangular area carries out width high etc. than extending what is obtained after 30-60 times.
(10) testing result stored in the corresponding testing result sequence of each expected areas is counted, confidence level is calculatedWherein signal represents one of green three kinds of testing results of reddish yellow,
(11) that color of selection maximum confidence is exported as final traffic lights testing result.
(12) output result passes to blind person by handset module with the form of voice.
3. method according to claim 2, it is characterised in that the SVM training patterns in the step 7, by with lower section
Method is obtained：
(1) coloured image Color of the color camera collection comprising green pavement traffic lights, is transmitted to compact processor and is processed.
(2) nonqualifying pixels extraction is carried out in HSV space.The span of HSV space triple channel be Hue ∈ [0,360),
Saturation ∈ [0,256), and Value ∈ [0,256), the nonqualifying pixels are to meet Value>Va also, and Hue<Hu1 or
Person Hue>The pixel of hu2；Wherein 50<va<256,250>hu1>230,250<hu2<300.
(3) to all nonqualifying pixels, connected region is extracted using region-growing method, each connected region is used as a candidate regions
Domain, calculates candidate region area a；
(4) for any one candidate region, the minimum rectangular area for covering the candidate region is set up, and the rectangular area is every
While parallel to the side of coloured image.And the height and width of the minimum rectangular area are extracted, the height for obtaining the minimum rectangular area is wide
Than r, and filling rate f (candidate region and minimum rectangular area area ratio)；Further candidate region (is connected by size a
Logical domain area), the screening of depth-width ratio r and filling rate f filters out a1<a<a2、r1<r<The qualified candidate regions of r2 and f ＞ f1
Domain, wherein, 10-6area<a1<10-4Area, 10-3area<a2<10-1area,0<r1<1,1.5<r2<4,0.2<1<0.6, wherein
Area is Color areas.
(5) it is extension center with the center of minimum rectangular area, three to six times is expanded to by high grade ratio wide to minimum rectangular area,
Obtain rectangular extension region.In calculating the extension rectangular area, the back of the body beyond the mean flow rate v1 of candidate region and candidate region
The mean flow rate v2 of scene area, filters out v1>The rectangular extension region of v2 is used as detection zone.
(6) detection zone that step 5 screening is obtained is carried out into gray processing treatment, and unified pixel size, extracts HOG or LBP figures
As feature.
(7) HOG the or LBP characteristics of image of yellow pavement traffic lights and rising star trade traffic lights is obtained according to step 1-6.
(8) data set for having marked is trained using SVM, kernel is LINEAR or RBF functions, is obtained using cross validation method
Optimal precision and recall rate.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710056650.1A CN106874945B (en) | 2017-01-25 | 2017-01-25 | Sidewalk traffic light detection system and method for visually impaired people |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710056650.1A CN106874945B (en) | 2017-01-25 | 2017-01-25 | Sidewalk traffic light detection system and method for visually impaired people |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN106874945A true CN106874945A (en) | 2017-06-20 |
| CN106874945B CN106874945B (en) | 2020-05-12 |
Family
ID=59158202
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710056650.1A Active CN106874945B (en) | 2017-01-25 | 2017-01-25 | Sidewalk traffic light detection system and method for visually impaired people |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN106874945B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113011251A (en) * | 2021-02-03 | 2021-06-22 | 深圳大学 | Pedestrian traffic light identification method based on traffic light geometric attributes |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN104299412A (en) * | 2014-09-29 | 2015-01-21 | 丁玉清 | Device assisting the blind in crossing road |
| JP2016501408A (en) * | 2012-12-03 | 2016-01-18 | コンティ テミック マイクロエレクトロニック ゲゼルシャフト ミットベシュレンクテル ハフツングＣｏｎｔｉ Ｔｅｍｉｃ ｍｉｃｒｏｅｌｅｃｔｒｏｎｉｃ ＧｍｂＨ | Method for supporting a signal phase assistant in a vehicle that recognizes signals |
| CN105913041A (en) * | 2016-04-27 | 2016-08-31 | 浙江工业大学 | Pre-marked signal lights based identification method |
| CN106203390A (en) * | 2016-07-22 | 2016-12-07 | 杭州视氪科技有限公司 | A kind of intelligent blind aid system |
| CN106214437A (en) * | 2016-07-22 | 2016-12-14 | 杭州视氪科技有限公司 | A kind of intelligent blind auxiliary eyeglasses |
-
2017
- 2017-01-25 CN CN201710056650.1A patent/CN106874945B/en active Active
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2016501408A (en) * | 2012-12-03 | 2016-01-18 | コンティ テミック マイクロエレクトロニック ゲゼルシャフト ミットベシュレンクテル ハフツングＣｏｎｔｉ Ｔｅｍｉｃ ｍｉｃｒｏｅｌｅｃｔｒｏｎｉｃ ＧｍｂＨ | Method for supporting a signal phase assistant in a vehicle that recognizes signals |
| CN104299412A (en) * | 2014-09-29 | 2015-01-21 | 丁玉清 | Device assisting the blind in crossing road |
| CN105913041A (en) * | 2016-04-27 | 2016-08-31 | 浙江工业大学 | Pre-marked signal lights based identification method |
| CN106203390A (en) * | 2016-07-22 | 2016-12-07 | 杭州视氪科技有限公司 | A kind of intelligent blind aid system |
| CN106214437A (en) * | 2016-07-22 | 2016-12-14 | 杭州视氪科技有限公司 | A kind of intelligent blind auxiliary eyeglasses |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113011251A (en) * | 2021-02-03 | 2021-06-22 | 深圳大学 | Pedestrian traffic light identification method based on traffic light geometric attributes |
| CN113011251B (en) * | 2021-02-03 | 2024-06-04 | 深圳大学 | Pedestrian traffic light identification method based on geometric attributes of traffic lights |
Also Published As
| Publication number | Publication date |
|---|---|
| CN106874945B (en) | 2020-05-12 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Son et al. | Real-time illumination invariant lane detection for lane departure warning system | |
| Varma et al. | Real time detection of speed hump/bump and distance estimation with deep learning using GPU and ZED stereo camera | |
| KR101833359B1 (en) | Method and apparatus for collecting traffic information from bigdata of outside image of car | |
| KR102133913B1 (en) | Digital display board system using object recognition based on deep learning for providing customized information of object and method for controlling thereof | |
| CN102509090B (en) | A kind of vehicle feature recognition device based on public safety video image in sky net engineering | |
| KR102122859B1 (en) | Method for tracking multi target in traffic image-monitoring-system | |
| US11423664B2 (en) | Warning device, warning method, and warning program | |
| CN106971544B (en) | A kind of direct method that vehicle congestion is detected using still image | |
| CN112368711A (en) | Method and apparatus for computer vision | |
| KR102122850B1 (en) | Solution for analysis road and recognition vehicle license plate employing deep-learning | |
| WO2020258077A1 (en) | Pedestrian detection method and device | |
| CN112200044B (en) | Abnormal behavior detection method and device and electronic equipment | |
| CN103996041A (en) | Vehicle color identification method and system based on matching | |
| Cabani et al. | Color-based detection of vehicle lights | |
| CN110378243A (en) | A kind of pedestrian detection method and device | |
| CN102867417A (en) | Taxi anti-forgery system and taxi anti-forgery method | |
| Vani et al. | Intelligent traffic control system with priority to emergency vehicles | |
| CN110415544A (en) | A kind of hazard weather method for early warning and automobile AR-HUD system | |
| CN108805060A (en) | A kind of zebra line style crossing detection method | |
| CN113066048A (en) | Segmentation map confidence determination method and device | |
| CN112313662A (en) | Method and apparatus for computer vision | |
| CN103324957A (en) | Identification method and identification device of state of signal lamps | |
| KR102499340B1 (en) | Hybrid video analysis device based on object filters and method | |
| CN106874945A (en) | A kind of pavement traffic lights detecting system and method for visually impaired people | |
| CN111062311B (en) | Pedestrian gesture recognition and interaction method based on depth-level separable convolution network |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| CB02 | Change of applicant information |
Address after: 9, 181, 310000, Wuchang Road, Wuchang Street, Yuhang District, Zhejiang, Hangzhou, 202-7 Applicant after: Hangzhou vision krypton Technology Co., Ltd. Address before: Room 589, C building, No. 525 Xixi Road, Xihu District, Zhejiang, Hangzhou 310007, China Applicant before: Hangzhou vision krypton Technology Co., Ltd. |
|
| GR01 | Patent grant | ||
| GR01 | Patent grant |