CN115752441B - Traffic light construction method for high-precision map - Google Patents
Traffic light construction method for high-precision map Download PDFInfo
- Publication number
- CN115752441B CN115752441B CN202211490521.0A CN202211490521A CN115752441B CN 115752441 B CN115752441 B CN 115752441B CN 202211490521 A CN202211490521 A CN 202211490521A CN 115752441 B CN115752441 B CN 115752441B
- Authority
- CN
- China
- Prior art keywords
- traffic light
- traffic
- light
- points
- image
- 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.)
- Active
Links
- 238000010276 construction Methods 0.000 title claims abstract description 24
- 230000035807 sensation Effects 0.000 claims abstract description 13
- 239000003086 colorant Substances 0.000 claims abstract description 12
- 238000002372 labelling Methods 0.000 claims description 8
- 238000005286 illumination Methods 0.000 claims description 7
- 238000012545 processing Methods 0.000 claims description 4
- 230000010354 integration Effects 0.000 claims description 3
- 239000013598 vector Substances 0.000 claims description 2
- 238000000034 method Methods 0.000 abstract description 7
- 238000005516 engineering process Methods 0.000 description 12
- 238000000605 extraction Methods 0.000 description 6
- 238000011161 development Methods 0.000 description 3
- 230000006870 function Effects 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000001360 synchronised effect Effects 0.000 description 2
- 238000013473 artificial intelligence Methods 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 230000001737 promoting effect Effects 0.000 description 1
- 238000009877 rendering Methods 0.000 description 1
- 229920006395 saturated elastomer Polymers 0.000 description 1
- 238000012216 screening Methods 0.000 description 1
- 230000003595 spectral effect Effects 0.000 description 1
Landscapes
- Traffic Control Systems (AREA)
Abstract
The traffic light construction method of the high-precision map provided by the invention has the advantages that the traffic light type is determined by carrying out light sensation identification, the RGB color space and the HSV color space are utilized for carrying out double judgment on the colors, and the identification accuracy is higher. Aiming at the traditional round traffic light, a scheme of drawing patterns at fixed points of circle centers is adopted, and the traffic light model can be built more accurately. For arrow type traffic lights, a contour point drawing scheme is adopted, so that the construction problem of a novel traffic light model is more accurately solved. For traffic lights with countdown, the method for acquiring the countdown data is used for synchronizing the traffic lights with the vehicle, so that the accuracy and the qualityof the data are ensured, and the occurrence of data error phenomenon is directly avoided. The invention is designed aiming at three traffic lights, so that the constructed high-precision map covers more comprehensive traffic light information.
Description
Technical Field
The invention belongs to the technical field of intelligent driving and map construction, and particularly relates to a technology for constructing traffic lights in a high-precision map.
Background
In the intelligent travel era, intelligent driving technology is widely accepted by the masses, intelligent driving is an important component of strategic emerging industry, and is the first highlight chapter in the process from the Internet era to the artificial intelligence era, and is one of the strategic high points of new economic and technological development in the world. The comprehensive development of the intelligent driving technology has great significance for promoting national technology, economy, society, life, safety and comprehensive national force, and provides great convenience for vast owners.
However, the intelligent driving technology is not perfect. Intelligent driving techniques rely mainly on sensors at the vehicle end to return information, which the vehicle uses to assist in determining vehicle surrounding and road information, thereby assisting the driver in driving. Combining with the development of the current age, automatic driving is gradually changed into a mainstream travel mode, and the technology which is mainly relied on by people for travel driving is used.
Traffic lights are the most important and frequently occurring elements in driving segments and have a relatively important impact on automatic driving. With the gradual perfection of intelligent driving technology, the demand of automatic driving on the construction precision of a high-precision map is gradually increased. The elements of the high-precision map are required to be more comprehensive and accurate. Among the many elements of the high-precision map, traffic lights appear very frequently, and traffic light elements are also very important for automatic driving. Therefore, how to construct perfect and accurate traffic light information in a high-precision map to meet the requirement of automatic driving to the greatest extent is very important.
The current traditional traffic light identification construction technology relies on the annotation identification of a sensing end, and the approximate shape and the annotation category of the red light are defined by BoundingBox frames. Although the traditional mode can construct a model of the red road lamp, the constructed model contains less information, and the intelligent driving technology can only learn that the road section has the traffic light from the figure. For the types of the red street lamps, the current red street lamp conditions and the like cannot be obtained, and if the high-precision map service automatic driving is better constructed, the red street lamp information needs to be contained as comprehensively as possible.
Disclosure of Invention
Aiming at the problems and needs existing in the prior art, the invention provides a traffic light construction method of a high-precision map, which is combined with SLAM technology (instant positioning and map construction technology) to construct a traffic light model so as to enable the traffic light model to contain more comprehensive information of the traffic light to meet the automatic driving requirement.
The technical scheme of the invention is as follows:
in order to achieve the above purpose, the invention provides a traffic light construction method of a high-precision map, which comprises the following steps:
S1: calibrating the position of a traffic light:
And (3) calibrating and identifying the air elements of the acquired data road section, determining the positions of the air elements, and marking the traffic light type, wherein the traffic light type refers to a unified general type of the traffic light.
S2: traffic light category determination by light sensation identification
S21: and selecting to collect data of the road section for n times under the set illumination condition, and ensuring that the collected data contains red light information and green light information. The set illumination condition is preferably in sunny day and night with good illumination.
S22: the shape of the current lighting is obtained through light sensation identification, and the traffic light category is judged according to the shape;
S3: sorting treatment according to different types of traffic lights
S31: round traffic lights.
S311: determining the circle center and the radius;
S312: determining the horizontal position of a traffic light;
S313, determining the position of the yellow lamp.
S32: arrow-shaped traffic light
S321: marking corner points;
s322: and (5) drawing an image.
S4: and obtaining countdown data for the traffic light with the countdown.
S5, information integration, and construction of complete traffic light model
And (3) integrating the information acquired in the steps S1 to S4, and constructing the position, the label, the type and the countdown data of the traffic lights together into a complete data set.
By adopting the technical scheme, the invention has the following advantages:
The invention is designed aiming at three traffic lights, so that the constructed high-precision map covers more comprehensive traffic light information.
The invention firstly carries out light sensation identification to determine the traffic light category, and carries out double judgment on the colors by utilizing the RGB color space and the HSV color space, so that the identification accuracy is higher.
Aiming at the traditional round traffic light, the invention adopts the scheme of drawing the pattern by the center of the circle at fixed points, and can construct the model of the traffic light more accurately.
The invention adopts a contour point drawing scheme for the arrow type traffic lights, and solves the construction problem of the novel traffic light model more accurately
The invention synchronizes the traffic lights with countdown by acquiring the countdown data, thus ensuring the accuracy and the qualitiness of the data and directly avoiding the occurrence of data error
Drawings
Fig. 1 is a flow chart of a traffic light construction method of a high-precision map.
Detailed Description
The traffic light construction method of the high-precision map is further clearly and completely described below by combining the drawings and the embodiments.
Firstly, the traffic light construction of the high-precision map is based on the following research and thinking:
there are currently three more common traffic light categories:
1. Traditional round traffic light
2. Arrow type traffic light
3. Traffic light with countdown function
The three traffic lights are the most common traffic light categories in the current driving road section, so the invention is mainly designed aiming at the three traffic lights, and the constructed high-precision map covers more comprehensive traffic light information. The invention provides different map construction schemes according to traffic light types.
For the traditional traffic lights and arrow-shaped traffic lights, the color is common, and the three colors of red, yellow and green are used for relevant representation although the display shapes of the traffic lights are different, so that when the construction schemes are respectively designed, a common processing scheme, namely the extraction of the light intensity, is adopted for the traditional traffic lights and the arrow-shaped traffic lights.
The three colors of red, yellow and green are different in numerical value identified by the light sensation identifier, the traditional light sensation extraction and identification uses an RGB color space, and the three basic colors of R (red), G (green) and B (blue) are used as bases to be overlapped in different degrees, so that rich and wide colors are generated. With this pattern over a thousand, six hundred thousand different colors can be identified. However, the three components of RGB have high correlation, and when one of the components is slightly changed, the corresponding color is changed, which has high requirements for accuracy of recognition.
Thus, the present invention introduces a new space: HSV color space, for color recognition. In this space, the parameters of the color are respectively: hue (H), saturation (S), brightness (V). The hue is in the range of 0 ° to 360 °,0 ° for red, 120 ° for green, and 60 ° for yellow as a complementary color, calculated counterclockwise from red. In terms of saturation, each color can be regarded as a mixed result of a certain spectral color and white, and the higher the saturation value is, the more saturated the color is, and the range of the value is 0% -100%. Brightness represents the brightness of the color, and normally black has a value of 0% and white has a value of 100%.
For the identification of the traffic lights, the invention utilizes the HSV color space to accurately identify and extract the lamp centers of each red light and each green light, and is not limited to BoundingBox position determination.
In addition, whether the traffic light is in a traditional round shape or an arrow-shaped traffic light, the lighting pattern in the current collection is obtained through extracting the color, and the lighting color is confirmed. Because the yellow lighting time is shorter and is not easy to collect, the red and green colors are generally collected, the three lamps of the traffic light are uniformly distributed and positioned in the same straight line, the interval distance between the geometric centers of the graphics of every two lamps is consistent, and the position of the yellow lamp can be accurately calculated after the positions of the red lamp and the green lamp are determined.
Further, aiming at the traditional round traffic lights, a scheme of determining the circle center and the radius is adopted, and because halation occurs in night luminosity extraction, larger errors can be generated in the extraction of the round radius, when data are acquired, the daytime with better illumination should be selected, after luminosity patterns are identified, the circle center and the radius are confirmed, the straight line where the traffic lights and the green lights are located is determined, the straight line formed by connecting the traffic lights and the green lights is used for calculating the center point, and the circle center of the yellow light can be obtained, wherein the radius of the yellow light is consistent with that of the traffic lights. If the circle center difference of the traffic lights is far smaller than the radius, the traffic lights can be judged to have only one display light, and the traffic lights are converted by changing colors.
For arrow-shaped traffic lights, the situation that the difference between the graph and the circle is large can be judged during light sensation identification, the corner points of the graph selected by the traffic lights are extracted, the corner points of the arrow are extracted as fully as possible, and the adjacent corner points are connected in a straight line, so that the graph is obtained.
At present, a plurality of traffic lights comprise countdown display boards, the number of the remaining seconds of the remaining red lights or green lights can be displayed, the number is continuously changed, if the difficulty is too high by utilizing image recognition and extraction, the difficulty of extracting and collecting data for many times is also too high, and therefore, the position and the size of the countdown board are determined only by adopting a BoundingBox method. For digital extraction, it can be obtained in a manner consistent with many current buses. The traffic light time synchronization function is installed in many buses at present to show at the rear of a vehicle display screen, be convenient for the rear of a vehicle when receiving bus sight obstruction, can utilize this relevant information of seeing preceding traffic light. The information is acquired by a distance limiting requirement, and the data can be acquired when the bus runs close to a traffic light for a certain distance. Therefore, the same method can be utilized when the map is constructed, when the vehicle runs close to the traffic light to a certain distance, the synchronous traffic light countdown data are automatically acquired and provided for the vehicle for intelligent driving, and the acquired data are very accurate and reliable.
Based on the above research, the invention provides a better traffic light construction method for a high-precision map.
Referring to fig. 1, one specific embodiment operates as follows:
s1: traffic light calibration by using traditional calibration technology
The sensing end firstly adopts a traditional calibration mode for the acquired data road section, performs unified calibration and identification for the air elements of the road section, and determines the positions of the air elements by BoundingBox (surrounding detection frame of target data). When labeled BoundingBox, four-dimensional vectors (x, y, w, h) are typically used for the image window to represent the center point coordinates and width and height of the window, respectively. For calibration images, the original Proposal is generally represented by a red box P, the target Ground Truth is represented by a green box G, and a relation is found such that the input original window P is mapped to obtain a regression window G.
In the calibration, the traffic light type is specially marked (i.e. a specific calibration value is given to the traffic light type, and all traffic light type data can be obtained directly by using the calibration value in the process of subsequent screening data), and the next processing is continued.
S2: traffic light category determination by light sensation identification
S21: and selecting to collect road section data for multiple times under the set illumination condition (in sunny days and nights with good illumination), and ensuring that the collected data contains red light information and green light information.
The colors are doubly judged by utilizing an RGB color space and an HSV color space, and the numerical requirement of red in the RGB color space is (225,0,0): the green value is required to be (0,225,0): in the HSV color space, the numerical requirement for red is hmin:0 or 156, hmax:10 or 180, smin:43, smax:255, vmin:46, vmax:255, the numerical requirements of green are: hmin:35, hmax:77, smin:43, smax:255, vmin:46, vmax:255, when judging, the HSV value sets the allowable error fluctuation to be up and down 3, and the current color can be identified only when the two color space values are in the specified range, and the current color is determined to be red or green.
S22: the general shape of the current lighting can be obtained through light sensation identification, the traffic light type can be judged according to the general shape, and the brightness identification result of the traditional round traffic light is round, so that the traffic light can be better identified and judged.
S3: and carrying out classification processing according to different types of traffic lights.
S31: for a traditional round traffic light, comprising:
s311: determining circle center and radius
Because the light sensation identification data acquired in the daytime when the light sensation is good does not contain halation, the image data are more accurate at the moment, and under the condition, the circle center and the radius of the acquired image are determined. Specifically, the obtained image is subjected to point cloud labeling, labeled points are classified into edge points and internal points, if distances from a certain point to all edge points in the internal are consistent, the internal point is determined to be a circle center, and the consistent distance is a circular radius. Red, red method for determining circle center radius by green light identical (consistent radius required).
When the distance between the centers of the traffic lights is smaller than the radius, the current condition is determined that the traffic lights only have one traffic light, and the traffic lights are changed by changing colors.
S312: determining the horizontal position of a traffic light
After the circle centers of the traffic lights and the green lights are determined, connecting the two circle centers to construct a straight line, wherein the straight line is the horizontal position of the current traffic lights.
S313, determining the position of the yellow light
And determining the midpoint of the connecting line of the circle centers of the traffic lights as the circle center position of the yellow light, judging whether the distance between the position and the circle center of the traffic lights is larger than the diameter length, and constructing the yellow light information when the condition is met, or else refusing to construct.
S32: for an arrow-shaped traffic light, comprising:
S321: corner point labeling
And (3) marking points on the edge points of the acquired image, wherein L marking is required to be carried out on the image points of the inflection point vertexes of the image, and one point can be marked by a distance from the rest points, so that the image marking is completed once.
S322: image rendering
And connecting marked points, namely connecting adjacent points, and drawing an image by using a straight line.
S4: obtaining countdown data
When the vehicle runs 500m away from the traffic light, the running vehicle starts to acquire the countdown information of the traffic light. And synchronized to the driving vehicle for intelligent driving.
S5, information integration, and construction of complete traffic light model
And finally, integrating the information acquired in the steps S1 to S4, and constructing the position, the label, the type and the countdown data of the traffic lights together into a complete data set, thus obtaining a complete traffic light model.
Claims (4)
1. The traffic light construction method of the high-precision map is characterized by comprising the following steps of:
S1: calibrating the position of a traffic light:
Calibrating and identifying the air elements of the acquired data road section, determining the positions of the air elements, and marking the traffic light type;
S2: traffic light category determination by light sensation identification
S21: selecting to collect data of the road section for n times under the set illumination condition, and ensuring that the collected data contains red light information and green light information;
s22: the shape of the current lighting is obtained through light sensation identification, and the traffic light category is judged according to the shape;
S3: processing according to different categories of traffic lights
S31: round traffic light:
S311: the circle center and the radius are determined, namely, the acquired image is subjected to point cloud labeling, labeled points are classified into edge points and internal points, if the distances from a certain point to all edge points in the internal part are consistent, the internal point is determined to be the circle center, the consistent distance is the circular radius, and when the distance of the circle center of a traffic light is smaller than the radius, the current condition is determined that the traffic light is only provided with a display lamp of one traffic light, and the change of the traffic light is performed by changing the color;
S312: determining the horizontal position of the traffic light, namely connecting two circle centers to construct a straight line, namely the horizontal position of the current traffic light;
s313, determining the position of the yellow light, namely determining the midpoint of a connecting line of the circle centers of the traffic lights as the position of the circle center of the yellow light, judging whether the distance between the position and the circle center of the traffic lights is larger than the diameter length, and constructing yellow light information if the condition is met, otherwise, not constructing the yellow light information;
s32: arrow-shaped traffic light
S321: the corner point labeling, namely, labeling points on the edge points of the acquired image, L labeling the image points on the corner points of the image, selecting a distance labeling point for the rest points, and completing the image labeling once;
S322: image drawing, namely connecting the marked adjacent points, and drawing the image by using straight lines;
s4: obtaining countdown data for a traffic light containing countdown;
s5, information integration, and construction of complete traffic light model
And (3) integrating the information acquired in the steps S1 to S4, and constructing the position, the type mark, the category and the countdown data of the traffic lights together into a complete data set.
2. The traffic light construction method of the high-precision map according to claim 1, wherein the step of S1 calibrating the traffic light position is to calibrate the position of an air element by BoundingBox;
when the window is calibrated, the image window is expressed by using four-dimensional vectors (x, y, w and h), and four values respectively represent the central point coordinates and the width and the height of the window; the calibration image uses a red box P to represent the original Proposal and a green box G to represent the target Ground Truth, so that the input original window P is mapped to a regression window G ̂ that is closer to the real window G.
3. The traffic light construction method according to claim 1, wherein in S21, the colors are doubly judged by using an RGB color space and an HSV color space, and the red or green color is determined when the values of the two color spaces meet the set range.
4. A traffic light construction method according to claim 3, wherein in S21, the numerical requirement of red in RGB color space is (225,0,0), and the numerical requirement of green is (0,225,0): in the HSV color space, the numerical requirement for red is hmin:0 or 156, hmax:10 or 180, smin:43, smax:255, vmin:46, vmax:255, the numerical requirements of green are: hmin:35, hmax:77, smin:43, smax:255, vmin:46, vmax:255.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211490521.0A CN115752441B (en) | 2022-11-25 | 2022-11-25 | Traffic light construction method for high-precision map |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211490521.0A CN115752441B (en) | 2022-11-25 | 2022-11-25 | Traffic light construction method for high-precision map |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN115752441A CN115752441A (en) | 2023-03-07 |
| CN115752441B true CN115752441B (en) | 2024-09-13 |
Family
ID=85337974
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211490521.0A Active CN115752441B (en) | 2022-11-25 | 2022-11-25 | Traffic light construction method for high-precision map |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115752441B (en) |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663892A (en) * | 2012-05-29 | 2012-09-12 | 四川川大智胜软件股份有限公司 | Method for correcting discoloration and enlarging of red light in night red-light-running vehicle picture |
| CN106446834A (en) * | 2016-09-27 | 2017-02-22 | 东软集团股份有限公司 | Vehicle type identification method and vehicle type identification device based on images |
Family Cites Families (8)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN109509255B (en) * | 2018-07-26 | 2022-08-30 | 京东方科技集团股份有限公司 | Tagged map construction and space map updating method and device |
| CN109377777A (en) * | 2018-12-21 | 2019-02-22 | 张奎 | A kind of additional transport lamp system and its control method |
| CN111830953B (en) * | 2019-04-12 | 2024-03-12 | 北京四维图新科技股份有限公司 | Vehicle self-positioning method, device and system |
| CN110197589B (en) * | 2019-05-29 | 2020-09-01 | 杭州诚道科技股份有限公司 | Deep learning-based red light violation detection method |
| CN112484739B (en) * | 2020-11-25 | 2023-05-30 | 中国第一汽车股份有限公司 | Map updating method, device, equipment and storage medium |
| CN112528918A (en) * | 2020-12-18 | 2021-03-19 | 浙江商汤科技开发有限公司 | Road element identification method, map marking method and device and vehicle |
| CN114475429B (en) * | 2022-02-21 | 2024-03-22 | 重庆长安汽车股份有限公司 | Traffic light reminding method and system combined with user driving intention and automobile |
| CN115346373A (en) * | 2022-08-16 | 2022-11-15 | 白犀牛智达(北京)科技有限公司 | Traffic light identification method and device |
-
2022
- 2022-11-25 CN CN202211490521.0A patent/CN115752441B/en active Active
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102663892A (en) * | 2012-05-29 | 2012-09-12 | 四川川大智胜软件股份有限公司 | Method for correcting discoloration and enlarging of red light in night red-light-running vehicle picture |
| CN106446834A (en) * | 2016-09-27 | 2017-02-22 | 东软集团股份有限公司 | Vehicle type identification method and vehicle type identification device based on images |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115752441A (en) | 2023-03-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110197589B (en) | Deep learning-based red light violation detection method | |
| Broggi et al. | Real time road signs recognition | |
| CN103345766B (en) | A kind of signal lamp recognition methods and device | |
| CN106909937B (en) | Traffic signal lamp identification method, vehicle control method and device and vehicle | |
| CN102176287B (en) | Traffic signal lamp identifying system and method | |
| CN105913041B (en) | It is a kind of based on the signal lamp recognition methods demarcated in advance | |
| CN103177256B (en) | Display state of traffic signal lamp recognition methods | |
| CN107016362B (en) | Vehicle weight recognition method and system based on vehicle front windshield pasted mark | |
| CN110688907B (en) | Method and device for identifying object based on night road light source | |
| CN102722707A (en) | License plate character segmentation method based on connected region and gap model | |
| CN105160691A (en) | Color histogram based vehicle body color identification method | |
| CN107506760A (en) | Traffic signals detection method and system based on GPS location and visual pattern processing | |
| CN104408424A (en) | Multiple signal lamp recognition method based on image processing | |
| CN102737221B (en) | Method and apparatus for vehicle color identification | |
| CN108256467B (en) | Traffic sign detection method based on visual attention mechanism and geometric features | |
| CN103324957B (en) | Identification method and identification device of state of signal lamps | |
| CN103093200A (en) | Algorithm for quickly and accurately locating plate number of image | |
| CN108446668A (en) | Traffic lights detection recognition method and system based on unmanned platform | |
| CN106919910A (en) | A kind of traffic sign recognition method based on HOG CTH assemblage characteristics | |
| CN102306276A (en) | Method for identifying color of vehicle body in video vehicle image based on block clustering | |
| CN112330961B (en) | Traffic indication red, yellow and green light image identification method based on time sequence rule | |
| CN110969647A (en) | Method for integrating vehicle identification tracking and vehicle lamp detection | |
| CN115752441B (en) | Traffic light construction method for high-precision map | |
| CN109800693B (en) | Night vehicle detection method based on color channel mixing characteristics | |
| CN108416284A (en) | A kind of dividing method of traffic lights |
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 |