WO2020220182A1 - 一种车道线检测方法、装置、控制设备及存储介质 - Google Patents

一种车道线检测方法、装置、控制设备及存储介质 Download PDF

Info

Publication number
WO2020220182A1
WO2020220182A1 PCT/CN2019/084934 CN2019084934W WO2020220182A1 WO 2020220182 A1 WO2020220182 A1 WO 2020220182A1 CN 2019084934 W CN2019084934 W CN 2019084934W WO 2020220182 A1 WO2020220182 A1 WO 2020220182A1
Authority
WO
WIPO (PCT)
Prior art keywords
lane line
initial
lane
local
target
Prior art date
Application number
PCT/CN2019/084934
Other languages
English (en)
French (fr)
Inventor
许睿
唐蔚博
陈竞
崔健
Original Assignee
深圳市大疆创新科技有限公司
Priority date (The priority date 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 date listed.)
Filing date
Publication date
Application filed by 深圳市大疆创新科技有限公司 filed Critical 深圳市大疆创新科技有限公司
Priority to CN201980004970.XA priority Critical patent/CN111295666A/zh
Priority to PCT/CN2019/084934 priority patent/WO2020220182A1/zh
Publication of WO2020220182A1 publication Critical patent/WO2020220182A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks

Definitions

  • the embodiment of the present invention relates to the field of control technology, and in particular to a lane line detection method, device, control device and storage medium.
  • the local map of the lane line is mainly used in the field of automatic driving. Based on the local map of the lane line, the driving plan of the current driving vehicle can be planned, and the establishment of the local map mainly depends on the detection of the lane line.
  • the current lane line detection methods usually need to rely on many assumptions (such as the plane assumption of the map, the parallel assumption of the lane line geometry, etc.) to complete the lane line detection. Such detection methods are used in some branch roads, urban areas, undulating roads, etc. It cannot work well in the scene, which makes the lane line detection result not accurate enough and not universal.
  • the embodiments of the present invention provide a lane line detection method, device, control device, and storage medium, which can be applied to a variety of lane line scenarios without relying too much on assumptions such as lane line geometrical parallel assumptions and road parallel assumptions.
  • an embodiment of the present invention provides a lane line detection method.
  • the method is applied to a control device, and there is a data connection between the control device and a movable platform.
  • the method includes:
  • Fitting optimization is performed on the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set to obtain the target lane line set.
  • an embodiment of the present invention provides a lane line detection device, which includes:
  • An acquisition module configured to acquire an environment image of the movable platform, and obtain a local map of the environment in which the movable platform is located according to the environment image, wherein the lane line position point is recorded in the local map;
  • a processing module configured to determine an initial lane line set from the local map, the initial lane line set including multiple initial lane line data;
  • the processing module is also used for fitting and optimizing the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set to obtain a target lane line set.
  • an embodiment of the present invention provides a control device, and a data connection exists between the control device and a movable platform.
  • the control device includes a vision sensor and a processor, and the processor is configured to:
  • Fitting optimization is performed on the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set to obtain the target lane line set.
  • the environment image collected by the control device is used to construct a local map where the movable platform is located, determine the initial lane line set based on the lane line position points in the local map, and compare the historical lane lines in the historical lane line set.
  • the data and the initial lane line data in the initial lane line set are fitted and optimized to obtain the target lane line set. It is possible to achieve lane line detection without relying on the assumption of geometric parallelism of lane lines and the assumption of road parallelism, which is conducive to better application to multiple vehicles.
  • FIG. 1 is a schematic flowchart of a method for detecting lane lines according to an embodiment of the present invention
  • FIG. 2 is a scene diagram of lane line detection provided by an embodiment of the present invention.
  • FIG. 3 is a schematic flowchart of a method for detecting lane lines according to another embodiment of the present invention.
  • FIG. 4 is a schematic flowchart of a lane line detection method provided by another embodiment of the present invention.
  • FIG. 5 is a schematic diagram of a connected domain label according to an embodiment of the present invention.
  • FIG. 6 is a schematic flowchart of a lane line detection method provided by another embodiment of the present invention.
  • FIG. 7 is a schematic diagram of an initial lane line set and a historical lane line set according to an embodiment of the present invention.
  • FIG. 8 is a schematic block diagram of a lane line detection device provided by an embodiment of the present invention.
  • Fig. 9 is a schematic block diagram of a control device according to an embodiment of the present invention.
  • an embodiment of the present invention proposes a multi-lane fusion lane line detection method, which can be collected in real time by visual sensors or according to a preset cycle.
  • the environment image of the mobile platform, and a local map of the environment where the mobile platform is located is constructed based on the environment image.
  • the preset optimization algorithm filters out the initial lane line from the set of local lane lines, and obtains the initial lane line set.
  • the initial lane line obtained by the filtering can be considered as the lane line most consistent with the actual lane line.
  • the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set can be fitted and optimized to obtain the target lane line set.
  • the current lane line detection result can be compared with the historical lane line detection result. The results are merged in time series and space to make the lane line detection results more accurate.
  • the movable platform may first determine the local map of the environment in which the movable platform is located, and connect based on the lane line position points in the local map. Domain analysis is used to obtain the local lane line set corresponding to the local map, and analyze the feature information of each local lane line in the local lane line set to obtain the weight value of each local lane line, and then based on the weighted maximum clique algorithm and each local The weight value of the lane line is optimized to determine at least one initial lane line to obtain the initial lane line set.
  • the initial lane line in the initial lane line set can be filtered to filter out the wrong lane line in the initial lane line, and the filtered initial lane line in the initial lane line set and the history in the historical lane line set
  • the lane line is fitted and optimized to obtain the target lane line set to complete the detection of the lane line.
  • the above-mentioned lane line detection method can be applied to a control device that has a data connection with a movable platform.
  • the movable platform may be some mobile devices that can drive on public transportation roads, such as autonomous vehicles;
  • the control device may be a driving auxiliary device that has a data connection with the movable platform, and the control device may be built in the movable
  • the platform such as a system in which the movable platform is integrated into the movable platform, can also be an external connection with the movable platform, such as an auxiliary driving device connected to the outside of the movable platform.
  • the lane line detection method can be applied to the lane line detection scene as shown in FIG. 2, where the movable platform S10 can be an unmanned car driving on a public transportation road, and the control device
  • the S100 is built in the movable platform S10, and one or more vision sensors S101 are also arranged on the movable platform S10, and the control device S100 can obtain the environment image of the movable platform 10 through the vision sensor S101.
  • the vision sensor S101 can be placed in the front, rear and/or roof of the movable platform S10, and one or more vision sensors S101 placed in the movable platform S10 can be placed at the same position or at Different positions are not limited in the embodiment of the present invention.
  • the visual sensor S101 of the movable platform S10 is installed in front of the movable platform, and is used to obtain the front image.
  • the lane line image point in the front view can be recognized based on the image model, and combined with the pose information of the movable platform S10 and the environment image, the lane line image point can be converted to the world The local map under the coordinate system to obtain the local map of the environment where the movable platform is located, and to determine the corresponding lane line position point of the lane line image point in the local map.
  • the initial lane line set can be determined from the local map according to the lane line location point, and the initial lane line data in the initial lane line set and the historical lane line data in the historical lane line set can be fitted for optimization, Get the target lane line set to complete the lane line detection.
  • the lane line detection method proposed by the embodiment of the present invention does not rely on the assumption of geometric parallelism of the lane lines and the assumption of parallel road surface.
  • the lane lines of the bifurcated road in FIG. 2 overlap and have different directions. If the traditional lane line detection strategy is used, for example, the parallelism of the lane line is used for detection, such a lane line scene cannot be detected correctly.
  • the initial lane line set can be determined from the local map, and the historical lane line in the historical lane line set The data and the initial lane line data in the initial lane line set are fitted and optimized to obtain the target lane line set, which can be better applied to multiple lane line scenarios (such as bifurcated roads, urban areas, undulating roads, etc.).
  • the versatility of lane line detection is provided.
  • the movable platform S10 and the control device S100 in FIG. 2 are merely examples for illustration.
  • the movable platform shown in FIG. 2 may also be a mobile controller installed on a mobile device, etc.
  • FIG. 2 It is only an example of the scenario involved in the embodiment of the present invention, which is mainly used to illustrate the principle of realizing lane line detection based on the control device in the embodiment of the present invention.
  • FIG. 3 is a schematic flowchart of a lane line detection method proposed by an embodiment of the present invention. The method is applied to a control device, and there is a data connection between the control device and a movable platform, wherein the method includes the following steps :
  • S301 Acquire an environment image of the movable platform, and obtain a local map of the environment in which the movable platform is located according to the environment image, wherein the lane line position point is recorded in the local map.
  • the above-mentioned lane line position point is determined by image analysis of the environment image.
  • the control device can identify the lane line image points in the environment image based on the image model, and obtain a local map of the environment based on the pose information of the movable platform and the environment image, and then determine the lane line image points based on the The location point of the lane line in the local map.
  • the control device can first obtain the environmental image of the environment in which the mobile platform is located in real time or according to a preset period through a visual sensor arranged outside the movable platform. Further, the control device can first target the environmental image through the image model. Preliminary detection of lane lines is performed to obtain lane line image points. The lane line image points are combined with the current position information of the movable platform and the environment image to convert to a local map of the world coordinate system, and the lane line image is determined according to the coordinate conversion Point at the location of the lane line in the local map.
  • a convolutional neural network Convolutional Neural Networks, CNN
  • the position information of the current position of the movable platform can be obtained by a visual inertial navigation model (such as VINS).
  • the control device may determine an initial lane line set from the local map in S302, and the initial lane line set includes multiple initial lane line data.
  • the initial lane line data included in the initial lane line set may be determined according to the lane line position points recorded in the local map.
  • the above-mentioned local map may be a grid map in the world coordinate system, the grid corresponding to the lane line position point is a lane line grid, and each lane line grid includes a semantic information, the semantic information
  • the grid used to characterize the lane line is the grid corresponding to the position point of the lane line.
  • the control device can analyze and process each lane line grid in the local map to obtain the local lane line set corresponding to the local map.
  • the local lane line set includes at least one local lane line, and then the local lane line collection An initial lane line set is determined in, and the initial lane line set includes at least one initial lane line data.
  • the control device determines the implementation manner of the initial lane line set from the local map, which may be based on the schematic flowchart shown in FIG. 4. Specifically, the control device can perform connected domain analysis processing on each lane line grid in the local map based on the semantic information of each lane line grid to obtain the local lane line set corresponding to the local map, and based on the preset optimization algorithm and local The weight value of each local lane line in the lane line set is optimized to determine the initial lane line to obtain the initial lane line set. Wherein, the weight value of the local lane line is determined according to the characteristic information of the lane line.
  • control device when the control device performs connected domain analysis processing on each lane line grid in the local map based on the semantic information of each lane line grid, it may first determine from the local map according to the semantic information of each lane line grid Each lane line grid is extracted, and the connected domain of each lane line grid is extracted, and the connected domain label of the lane line grid in a small area is determined according to the semantic information of each lane line grid, based on the corresponding connectivity of each lane line grid The domain label can be used for lane line fitting to determine one or more local lane lines corresponding to the location points of the lane line in a small area, and so on, to determine all the local lane lines in the local map, and obtain the corresponding local map Local lane line assembly.
  • the lane line grids belonging to the same lane line correspond to the same connected domain label
  • the connected domain label is associated with the semantic information of the lane line grid.
  • the semantic information of the lane line grid G1 is used to indicate that the image point corresponding to the lane line grid G1 belongs to the lane line A
  • the semantic information of the lane line grid G2 is used to indicate that the image point corresponding to the lane line grid G2 is also Belonging to lane line A
  • the lane line grid G1 and lane line grid G2 correspond to the same connected domain label.
  • each grid represents an image point
  • the number in the grid represents the connected component label corresponding to each grid.
  • a grid with a domain label of 0 may be a grid of local lane lines that does not belong to a lane line
  • a grid with a connected domain label of not 0 may be a grid of lane lines that may belong to a lane line.
  • the grid with the connected domain label of 1 is the image point belonging to the A lane line in the local map
  • the image point with the connected domain label 2 is the image point corresponding to the B lane line in the local map, where the A lane line and the B
  • the lane lines are two different lane lines.
  • control device when it performs connected domain analysis processing on the partial map, it can select a small-scale grid from the partial map according to an image detection window of a preset size, and determine based on the semantic information of each grid
  • the connected component label of each grid in the image detection window wherein the connected component label corresponding to the non-lane line grid is 0, and the connected component label of the lane line grid is determined by corresponding semantic information. Since the grids in the same connected component correspond to the same connected component label, the connected components of the image detection window can be determined based on the characteristics of whether the connected component labels are the same, that is, the grids included in each connected component are determined.
  • the preset size of the image detection window may be 3*3 or 5*5, for example.
  • the grid corresponding to each connected domain can be fitted according to the optimal solution algorithm to obtain the local lane line corresponding to each connected domain, and so on, the image detection can be determined
  • the local lane lines corresponding to the lane lines included when the window is located at different positions in the local map can determine all the local lane lines in the local map to obtain a set of local lane lines.
  • control device determines the local lane line set corresponding to the local map, it can optimize and determine at least one initial lane line based on the weighted maximum clique algorithm and the weight value of each local lane line in the local lane line set. At least one initial lane line is filtered to obtain an initial lane line set.
  • the weight value of the local lane line is determined according to the feature information of the lane line.
  • the feature information of the lane line includes: the geometric feature and/or color feature of the lane line, and the geometric feature includes the length feature and the width feature. And any one or more of the parallel features between lane lines.
  • the corresponding relationship between the geometric and color features of the lane line and the hypothetical score can be established in advance.
  • the corresponding relationship between the geometric feature, the color feature and the hypothetical score of the lane line may be as shown in Table 1. It can be seen that each local lane line can determine its own corresponding hypothetical score sum through the corresponding relationship shown in Table 1, and the hypothetical score sum is the weight value corresponding to the local lane line.
  • the hypothetical score corresponding to each feature dimension in Table 1 is only an example, which is mainly used to illustrate the principle of determining the weight value of a local lane line through the feature information of the lane line, and cannot be a limitation to the embodiment of the present invention.
  • control device may analyze the characteristic information of each local lane line, and determine the weight value of each local lane line according to the pre-established correspondence between the geometric characteristics, color characteristics and the assumed score of the lane line. Further, based on the weighted maximum clique algorithm, at least one local lane line with the highest weight value among the local lane lines is solved, that is, at least one initial lane line.
  • control device may perform filtering processing on at least one initial lane line to obtain an initial lane line set.
  • post-processing may be performed based on the preset a priori information for the initial lane line, so as to filter all the initial lane lines of the at least one initial lane line, and the wrong lanes in all the initial lane lines Lines are filtered out to get the initial lane line set.
  • the preset a priori information for the initial lane line is a preset range or a preset value set based on the national standard of the lane line, and the a priori information specifically includes length information, width information, etc., assuming that the country The standard lane line length is 1.5 meters, then the length information included in the a priori information for the initial lane line can be set to be a part of the length less than the standard lane line length, such as the range of 10 cm to 15 cm, etc.; assuming national standards If the width of the lane line is 15 cm, the width information included in the prior information can be set to a range of 13 cm to 17 cm. Correspondingly, the wrong lane line is a lane line that does not meet the preset prior information.
  • the control device After the control device determines the initial lane line set from the local map, the control device can perform fitting optimization on the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set in S303 to obtain the target Lane line collection.
  • the current lane line detection results can be combined with the historical lane line detection results in time sequence and space, which is beneficial to improve the accuracy of the lane line detection results.
  • the control device can query whether there is a historical lane line set in the storage area. If the historical lane line set does not exist, it can determine the current lane line detection It is the first lane line detection, and the initial lane line set obtained this time is stored in the storage area as the historical lane line set to facilitate subsequent fitting optimization with the new initial lane line set. If there is a set of historical lane lines, the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set can be fitted and optimized to obtain the target lane line set.
  • the method of constructing the overall lane line starting from the local map of the environment where the movable platform is located in the embodiment of the present invention can be used on the one hand. To a certain extent, improve the accuracy of lane line fitting and improve the accuracy of lane line recognition. On the other hand, because the area occupied by the local map is relatively small, it can also save the movable platform to a certain extent when performing lane line fitting. speed.
  • the control device can obtain the environment image of the movable platform, and obtain the local map of the environment where the movable platform is located according to the environment image, determine the initial lane line set from the local map, and compare the historical lane line set
  • the historical lane line data and the initial lane line data in the initial lane line set are fitted and optimized to obtain the target lane line set. It is possible to realize the detection of lane lines without relying on the assumption of geometric parallelism of lane lines and the assumption of parallel road surface, which is beneficial to better application to multiple lane line scenarios and improves versatility.
  • FIG. 6 is a schematic flowchart of another lane line detection method proposed by an embodiment of the present invention.
  • the method is applied to a control device, and there is a data connection between the control device and a movable platform, wherein the method includes the following step:
  • S601 Acquire an environment image of the movable platform, and obtain a local map of the environment in which the movable platform is located according to the environment image, where the lane line position point is recorded in the local map. Further, the control device may determine an initial lane line set from the local map in S602, and the initial lane line set includes multiple initial lane line data. Further, after determining the initial lane line set from the local map, the control device may determine the target initial lane line from the initial lane line set in S603, and determine from the historical lane line set to match the target initial lane line The target historical lane line of, the target initial lane line is any one of the initial lane line set.
  • the target initial lane line matches the index information corresponding to the target historical lane line.
  • the local map may be a grid map in the world coordinate system, the grid corresponding to the lane line position point is a lane line grid, and each lane line grid includes a semantic information, and the semantic information is used
  • the grid representing the lane line is the grid corresponding to the position point of the lane line.
  • each initial lane line in the initial lane line set is curve-fitted by at least one lane line grid, it can be based on the composition
  • the voice information of the lane line grid of the initial lane line determines the lane to which each initial lane line belongs, and then adds index information to each initial lane line, and each index information indicates the lane to which the initial lane line belongs.
  • any initial lane line can be selected from the initial lane line set as the target lane line, and the index information of the target lane line can be compared with the index information corresponding to each historical lane line in the historical lane line set, if the comparison is obtained
  • the index information of the target lane line is the same as the index information corresponding to any historical lane line in the historical lane line set, then it can be determined that the index information of the target lane line matches the index information corresponding to any historical lane line in the historical lane line set , And determine the any historical lane line as the target historical lane line.
  • control device can select the next target lane line from the initial lane line set, and again detect whether there is a target historical lane line matching the next target lane line in the historical lane line set, and so on, until After detecting all the initial lane lines in the initial lane line set, it ends.
  • the initial lane line set determined by the control device from the local map is shown as 60 in the figure.
  • the initial lane line set 60 includes the initial lane line 1 and the initial lane line 2.
  • the index information of 1 indicates that the initial lane line 1 belongs to lane line A, and the index information of initial lane line 2 indicates that the initial lane line 2 belongs to lane line B;
  • the historical lane line set is shown as 61 in the figure, the historical lane line set Including historical lane line 1 and historical lane line 2.
  • the index information of historical lane line 1 indicates that historical lane line 1 belongs to lane line A
  • the index information of historical lane line 2 indicates that historical lane line 2 belongs to lane line B.
  • the control device can compare the respective index information corresponding to the initial lane line 1 and the initial lane line 2 with the index information corresponding to the historical lane line 1 and the historical lane line 2.
  • the index information of lane line 1 indicates that it belongs to lane line A. It can be determined that the initial lane line 1 matches the historical lane line 1, that is, the historical lane line 1 is the target historical lane line corresponding to the initial lane line 1; because the initial lane line 2
  • the index information of the historical lane line 2 and the historical lane line 2 both indicate that they belong to the lane line B, and it can be determined that the initial lane line 2 matches the historical lane line 2, that is, the historical lane line 2 is the target historical lane line corresponding to the initial lane line 2.
  • the initial lane line corresponding to the target initial lane line may be determined according to the over-fitting constraint condition and the parallel constraint condition.
  • the data and historical lane line data corresponding to the target historical lane line are fitted and optimized to obtain the target lane line, and at least one target lane line forms a target lane line set.
  • the initial lane line data corresponding to the target initial lane line and the historical lane line data corresponding to the target historical lane line can be fitted and optimized according to the lane model created by the over-fitting constraint condition and the parallel constraint condition to obtain the target Lane line.
  • each lane line can adopt an independent lane model, which is mainly used to fit and optimize the initial lane line data corresponding to the target initial lane line and the historical lane line data corresponding to the target historical lane line to obtain the target lane line .
  • the process of generating the target lane line by the lane model is essentially a process of continuously compressing the new target initial lane line to the target historical lane line, so as to achieve the fusion of the target initial lane line and the target historical lane line.
  • the mathematical function corresponding to the lane model can be composed of three parts.
  • the first part is the least squares model based on solving the target initial lane line corresponding to the lane line observation point cloud and the target historical lane line as the corresponding curve The minimum error between the equations;
  • the second part is the smoothing term of the curve (corresponding to the preset conditions of overfitting).
  • the three parts are weakly parallel constraints (corresponding to parallel constraints), which are mainly used to constrain the parallel state between two adjacent lanes.
  • the initial lane line set determined by the control device from the local map is shown as 60 in the figure
  • the historical lane line set is shown as 61 in the figure
  • the control device determines that the historical lane line 1 is The original lane line 1 corresponds to the target historical lane line
  • the historical lane line 2 is the target historical lane line corresponding to the initial lane line 2.
  • the control device can align the initial lane line 1 with the historical lane line 1 based on the lane model, and merge the initial lane line 1 and the historical lane line 1 to obtain the target lane line 621; based on the lane model, the initial lane line 2 Align with the historical lane line 2 and merge the initial lane line 2 and the historical lane line 2 to obtain the target lane line 622. Further, all target lane lines (ie, target lane line 621 and target lane line 622) form a target lane line set 62.
  • each lane line adopts an independent lane model to fit the target initial lane line and the target historical lane line, and the constraint of parallel state is added between two adjacent lane lines. Therefore, for complex lane detection scenes (such as branch road scenes, urban scenes, etc.), the accuracy of lane line detection can be guaranteed, so that it can be better applied to multiple lane line scenes and improve versatility.
  • complex lane detection scenes such as branch road scenes, urban scenes, etc.
  • the control device may combine the target lane lines in the target lane line set based on the attribute information of the lanes , Get at least one lane, and generate the lane centerline of the lane to facilitate the driving of the movable platform.
  • the attribute information includes geometric characteristics and/or color characteristics of the lanes, and the geometric characteristics include any one or more of length characteristics, width characteristics, and parallel characteristics between lanes.
  • the control device can obtain the environment image of the movable platform, and obtain a local map of the environment in which the movable platform is located according to the environment image, and determine the initial lane line set from the local map. Further, the target initial lane line is determined from the initial lane line set, and the target historical lane line matching the target initial lane line is determined from the historical lane line set, and then the target historical lane line is determined according to the over-fitting constraint and the parallel constraint The initial lane line data corresponding to the target initial lane line and the historical lane line data corresponding to the target historical lane line are fitted and optimized to obtain the target lane line. By increasing the parallel constraint conditions between two adjacent lanes, it is beneficial to better adapt to a variety of lane line scenarios and improve versatility.
  • An embodiment of the present invention provides a lane line detection device, which is used to execute a module of any one of the foregoing methods. Specifically, referring to FIG. 8, it is a lane line provided by an embodiment of the present invention.
  • the lane line detection device of this embodiment can be configured in a control device, which can have a data connection with a movable platform such as an autonomous vehicle.
  • the lane line detection device includes: an acquisition module 80 and Processing module 81.
  • the acquiring module 80 is used to acquire the environment image of the movable platform.
  • the processing module 81 is configured to obtain a local map of the environment in which the movable platform is located according to the environment image, wherein the lane line position points are recorded in the local map; the initial lane line is determined from the local map
  • the initial lane line set includes a plurality of initial lane line data; the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set are fitted and optimized to obtain the target lane line set .
  • the lane line position point is determined by performing image analysis on the environment image.
  • the initial lane line data included in the initial lane line set is determined according to the lane line position points recorded in the local map.
  • the processing module 81 is specifically configured to identify the lane line image points in the environment image based on the image model;
  • the local map is a grid map in the world coordinate system
  • the grid corresponding to the lane line position point is a lane line grid
  • each lane line grid includes a piece of semantic information. Semantic information is used to represent that the lane line grid is a grid corresponding to the lane line position point.
  • the processing module 81 is specifically configured to analyze and process each lane line grid in the local map to obtain a local lane line set corresponding to the local map, and the local lane line set includes at least one Local lane line; the initial lane line set is determined from the set of local lane lines.
  • the processing module 81 is specifically configured to perform connected domain analysis processing on each lane line grid in the local map based on the semantic information of each lane line grid to obtain the local lane line corresponding to the local map set.
  • the processing module 81 is specifically configured to optimally determine the initial lane line based on a preset optimization algorithm and the weight value of each local lane line in the local lane line set to obtain the initial lane line set; wherein, The weight value of the local lane line is determined according to the characteristic information of the lane line.
  • the processing module 81 is specifically configured to optimally determine at least one initial lane line based on the weighted maximum clique algorithm and the weight value of each local lane line in the set of local lane lines; The lane lines are filtered to obtain the initial lane line set.
  • the characteristic information of the lane line includes: geometric characteristics and/or color characteristics of the lane line, and the geometric characteristics include any one of length characteristics, width characteristics, and parallel characteristics between lane lines. Kind or more.
  • the processing module 81 is specifically configured to determine a target initial lane line from the set of initial lane lines, and determine a target historical lane line matching the target initial lane line from the set of historical lane lines ,
  • the target initial lane line is any one of the initial lane line set;
  • the initial lane line data corresponding to the target initial lane line and the historical lane line data corresponding to the target historical lane line are fitted and optimized to obtain the target lane line;
  • the target lane line composes the target lane line set.
  • the target initial lane line matches the index information corresponding to the target historical lane line.
  • the processing module 81 is further configured to combine the target lane lines in the target lane line set based on the attribute information of the lanes to obtain at least one lane, and generate the lane center line of the lane;
  • the attribute information includes geometric characteristics and/or color characteristics of the lanes, and the geometric characteristics include any one or more of length characteristics, width characteristics, and parallel characteristics between lanes.
  • the lane line detection device provided in this embodiment can execute the lane line detection method as shown in FIG. 3 and FIG. 6 provided in the foregoing embodiment, and the execution method and beneficial effects are similar, and will not be repeated here.
  • FIG. 9 is a structural diagram of a control device provided by an embodiment of the present invention, as shown in FIG. 9
  • the control device 90 includes a memory 901, a processor 902, and a visual sensor 903.
  • the processor 902 may be a central processing unit (CPU).
  • the processor 902 may be a hardware chip.
  • the hardware chip may be an application-specific integrated circuit (ASIC), a programmable logic device (PLD) or a combination thereof.
  • the PLD may be a complex programmable logic device (CPLD), a field-programmable gate array (FPGA), a general array logic (generic array logic, GAL) or any combination thereof.
  • a program code is stored in the memory 901, and the processor 902 calls the program code in the memory.
  • the processor 902 performs the following operations: obtains the environment image of the movable platform through the visual sensor 903, and according to the The environment image obtains a local map of the environment in which the movable platform is located, wherein the lane line position points are recorded in the local map;
  • Fitting optimization is performed on the historical lane line data in the historical lane line set and the initial lane line data in the initial lane line set to obtain the target lane line set.
  • the lane line position point is determined by performing image analysis on the environment image.
  • the initial lane line data included in the initial lane line set is determined according to the lane line position points recorded in the local map.
  • the processor 902 is specifically configured to recognize lane line image points in the environment image based on an image model
  • the local map is a grid map in the world coordinate system
  • the grid corresponding to the lane line position point is a lane line grid
  • each lane line grid includes a piece of semantic information. Semantic information is used to represent that the lane line grid is a grid corresponding to the lane line position point.
  • the processor 902 is specifically configured to analyze and process each lane line grid in the local map to obtain a local lane line set corresponding to the local map, and the local lane line set includes at least one Local lane line; the initial lane line set is determined from the set of local lane lines.
  • the processor 902 is specifically configured to perform connected domain analysis processing on each lane line grid in the local map based on the semantic information of each lane line grid to obtain the local lane line corresponding to the local map set.
  • the processor 902 is specifically configured to optimize and determine the initial lane line based on a preset optimization algorithm and the weight value of each local lane line in the set of local lane lines to obtain the initial lane line set; wherein, The weight value of the local lane line is determined according to the characteristic information of the lane line.
  • the processor 902 is specifically configured to optimally determine at least one initial lane line based on the weighted maximum clique algorithm and the weight value of each local lane line in the set of local lane lines; The lane lines are filtered to obtain the initial lane line set.
  • the characteristic information of the lane line includes: geometric characteristics and/or color characteristics of the lane line, and the geometric characteristics include any one of length characteristics, width characteristics, and parallel characteristics between lane lines. Kind or more.
  • the processor 902 is specifically configured to determine a target initial lane line from the set of initial lane lines, and determine a target historical lane line matching the target initial lane line from the set of historical lane lines ,
  • the target initial lane line is any one of the initial lane line set;
  • the initial lane line data corresponding to the target initial lane line and the historical lane line data corresponding to the target historical lane line are fitted and optimized to obtain the target lane line;
  • the target lane line composes the target lane line set.
  • the target initial lane line matches the index information corresponding to the target historical lane line.
  • the processor 902 is further configured to combine the target lane lines in the target lane line set based on the attribute information of the lanes to obtain at least one lane, and generate the lane center line of the lane;
  • the attribute information includes geometric characteristics and/or color characteristics of the lanes, and the geometric characteristics include any one or more of length characteristics, width characteristics, and parallel characteristics between lanes.
  • the movable platform provided in this embodiment can execute the lane line detection method as shown in FIG. 3 and FIG. 6 provided in the foregoing embodiment, and the execution mode and beneficial effects are similar, and will not be repeated here.
  • the embodiment of the present invention also provides a computer program product containing instructions, which when run on a computer, causes the computer to execute the relevant steps of the lane line detection method described in the foregoing method embodiment.
  • the program can be stored in a computer readable storage medium. During execution, it may include the procedures of the above-mentioned method embodiments.
  • the storage medium may be a magnetic disk, an optical disc, a read-only memory (Read-Only Memory, ROM), or a random access memory (Random Access Memory, RAM), etc.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Artificial Intelligence (AREA)
  • Molecular Biology (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Multimedia (AREA)
  • Traffic Control Systems (AREA)
  • Image Analysis (AREA)

Abstract

本发明实施例提供了一种车道线检测方法、装置、控制设备及存储介质，该方法包括：采集的环境图像构造可移动平台所处的局部地图，基于局部地图中车道线位置点确定出初始车道线集合，并对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合，可以不依赖于车道线几何平行假设和路面平行假设，实现车道线检测，有利于更好地适用于多种车道线场景，提高通用性。

Description

一种车道线检测方法、装置、控制设备及存储介质 技术领域
本发明实施例涉及控制技术领域，尤其涉及一种车道线检测方法、装置、控制设备及存储介质。
背景技术
车道线局部地图主要应用于自动驾驶领域，基于车道线局部地图可以规划当前行驶车辆的行驶计划，而局部地图的建立主要依赖于对车道线的检测。当前采用的车道线检测方法通常需要依赖诸多假设(如地图的平面假设、车道线几何结构的平行假设等)来完成车道线的检测，这样的检测方式，在一些分岔路、城区、起伏路面等场景下无法很好地工作，使得车道线检测结果不够准确，不具有通用性。
发明内容
有鉴于此，本发明实施例提供了一种车道线检测方法、装置、控制设备及存储介质，可适用于多种车道线场景而不过于依赖车道线几何平行假设和路面平行假设等假设条件。
一方面，本发明实施例提供了一种车道线检测方法，所述方法应用于控制设备，所述控制设备与可移动平台之间存在数据连接，该方法包括：
获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
另一方面，本发明实施例提供了一种车道线检测装置，该装置包括：
获取模块，用于获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
处理模块，用于从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
所述处理模块，还用于对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
再一方面，本发明实施例提供了一种控制设备，所述控制设备与可移动平台之间存在数据连接，该控制设备包括：视觉传感器和处理器，所述处理器用于：
通过所述视觉传感器获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
在本发明实施例中，控制设备可以采集的环境图像构造可移动平台所处的局部地图，基于局部地图中车道线位置点确定出初始车道线集合，并对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合，可以不依赖于车道线几何平行假设和路面平行假设，实现车道线检测，有利于更好地适用于多种车道线场景，提高通用性。
附图说明
为了更清楚地说明本发明实施例或现有技术中的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，显而易见地，下面描述中的附图仅仅是本发明的一些实施例，对于本领域普通技术人员来讲，在不付出创造性劳动的前提下，还可以根据这些附图获得其他的附图。
图1是本发明实施例提供的一种车道线检测方法的示意流程图；
图2为本发明实施例提供的一种车道线检测场景图；
图3为本发明另一实施例提供的一种车道线检测方法的示意流程图；
图4为本发明又一实施例提供的一种车道线检测方法的示意流程图；
图5为本发明实施例提供的一种连通域标签的示意图；
图6为本发明又一实施例提供的一种车道线检测方法的示意流程图；
图7为本发明实施例的提供的一种初始车道线集合和历史车道线集合的示意图；
图8为本发明实施例提供的一种车道线检测装置的示意性框图；
图9是本发明实施例提供的一种控制设备的示意性框图。
具体实施方式
为了使得车道线检测适用于更多的场景，提高车道线检测的通用性，本发明实施例提出了一种多车道融合的车道线检测方法，该方法通过视觉传感器实时或者按照预设周期采集可移动平台的环境图像，并基于该环境图像构造可移动平台所处环境的局部地图。进一步地，可以基于局部地图中的车道线位置点进行车道线拟合得到局部车道线集合，并基于车道线的特征信息确定各局部车道线的权重值，进而基于各局部车道线的权重值和预设优化算法从该局部车道线集合中筛选出初始车道线，得到初始车道线集合，筛选得到的初始车道线可以认为是和实际车道线最相符的车道线。进一步地，可以对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合，可以将当前的车道线检测结果与历史车道线检测结果进行时序和空间上的融合，使得车道线检测结果更加准确。
在一个实施例中，可参见如图1所示的车道线检测方法的示意流程图，可移动平台可先确定可移动平台所处环境的局部地图，基于局部地图中的车道线位置点进行连通域分析，得到局部地图对应的局部车道线集合，并对局部车道线集合中各局部车道线的特征信息进行分析，以得到各局部车道线的权重值，进而基于带权最大团算法和各局部车道线的权重值，优化确定出至少一条初始车道线，得到初始车道线集合。进一步地，可以对初始车道线集合中的初始车道线进行过滤处理，滤除初始车道线中的错误车道线，并将初始车道线集合中 过滤后的初始车道线与历史车道线集合中的历史车道线，进行拟合优化，得到目标车道线集合，以完成对车道线的检测。
在一个实施例中，上述车道线检测方法可以应用于控制设备，该控制设备与可移动平台存在数据连接。其中，该可移动平台可以为一些能够行驶在公共交通道路上的移动装置，例如自动驾驶车辆；该控制设备可以为与可移动平台存在数据连接的辅助驾驶装置，该控制设备可以内置于可移动平台，如可移动平台整合于可移动平台中的系统等，也可以为与可移动平台存在外部连接，如连接于可移动平台外部的辅助驾驶装置。
在一个实施例中，所述车道线检测方法可应用在如图2所示的车道线检测场景中，其中，可移动平台S10可以一辆行驶在公共交通道路上的无人驾驶汽车，控制设备S100内置于可移动平台S10，该可移动平台S10上还安置有一个或者多个视觉传感器S101，控制设备S100可以通过视觉传感器S101获取可移动平台10的环境图像。其中，所述视觉传感器S101可安置在可移动平台S10的前方、后方和/或车顶等位置，可移动平台S10中安置的一个或多个视觉传感器S101可以安置在相同位置，也可安置在不同位置，在本发明实施例中不做限定。
通常，可移动平台S10的视觉传感器S101安装在可移动平台的前方，用于获取前方的图像。在获取了前方的图像，或称为前视图后，可基于图像模型识别前视图中的车道线图像点，并结合可移动平台S10的位姿信息和环境图像将该车道线图像点转换到世界坐标系下的局部地图，以得到可移动平台所处环境的局部地图，并确定该车道线图像点在局部地图中对应的车道线位置点。进一步地，可以根据该车道线位置点从局部地图中确定出初始车道线集合，并将初始车道线集合中的初始车道线数据与历史车道线集合中的历史车道线数据，进行拟合优化，得到目标车道线集合，以完成对车道线的检测。
可以看出，本发明实施例提出的车道线检测方法，不依赖于车道线几何平行假设和路面平行假设，例如图2中的分岔路口的车道线存在交叠且方向不一致的情况，此时如果使用传统的车道线检测策略例如利用车道线的平行性进行检测，则无法正确地检测出这种车道线场景。
因此，为了更好地检测一些非常规性的车道线问题，提供一种更通用的车 道线检测方法，可以从局部地图中确定出初始车道线集合，并对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合，从而能够更好地适用于多种车道线场景(如分岔路、城区，起伏路面等场景)，提升了车道线检测的通用性。
其中，图2中的可移动平台S10和控制设备S100仅为举例说明，在其他例子中，图2所示的可移动平台还可以为安装在移动装置上的移动控制器等，同时，图2也仅为对本发明实施例所涉及的场景进行举例，主要用于说明本发明实施例的基于控制设备实现车道线检测的原理。
请参见图3，是本发明实施例提出的一种车道线检测方法的流程示意图，该方法应用于控制设备，该控制设备与可移动平台之间存在数据连接，其中，所述方法包括以下步骤：
S301，获取可移动平台的环境图像，并根据环境图像得到可移动平台所处环境的局部地图，其中，在局部地图中记录了车道线位置点。
其中，上述车道线位置点是通过对环境图像进行图像分析确定的。在一个实施例中，控制设备可以基于图像模型识别环境图像中的车道线图像点，并基于可移动平台的位姿信息和环境图像得到所处环境的局部地图，进而根据车道线图像点确定出局部地图中的车道线位置点。
作为一种可行的实施方式，控制设备可先通过设置在可移动平台外的视觉传感器实时或者依照预设周期获取可移动平台所处环境的环境图像，进一步地，可先通过图像模型针对环境图像作关于车道线的初步检测，得到车道线图像点，通过将车道线图像点结合可移动平台当前的位置信息和环境图像转换到世界坐标系的局部地图，并根据坐标转换，确定出车道线图像点在局部地图中的车道线位置点。在一个实施例中，可采用卷积神经网络(Convolutional Neural Networks，CNN)对环境图像作初步的车道线检测。在一个实施例中，可移动平台当前位置的位置信息可以由视觉惯导模型(如VINS)获取。
在得到可移动平台所处环境的局部地图之后，控制设备可以在S302中从局部地图中确定出初始车道线集合，该初始车道线集合包括多个初始车道线数据。
其中，初始车道线集合中包括的初始车道线数据可以是根据局部地图中记录的车道线位置点确定的。在一个实施例中，上述局部地图可以为在世界坐标系下的网格地图，该车道线位置点对应的网格为车道线网格，每个车道线网格包括一个语义信息，该语义信息用于表征车道线网格为车道线位置点对应的网格。针对这种情况，控制设备可以对局部地图中的各个车道线网格进行分析处理，得到局部地图对应的局部车道线集合，该局部车道线集合包括至少一个局部车道线，进而从局部车道线集合中确定出初始车道线集合，该初始车道线集合包括至少一个初始车道线数据。
在一个实施例中，控制设备从局部地图中确定出初始车道线集合的实现方式，可基于图4所示的流程示意图。具体地，控制设备可以基于各个车道线网格的语义信息对局部地图中的各个车道线网格进行连通域分析处理，得到局部地图对应的局部车道线集合，并基于预设的优化算法和局部车道线集合中各局部车道线的权重值，优化确定出初始车道线，得到初始车道线集合。其中，该局部车道线的权重值是根据车道线的特征信息确定的。
在一个实施例中，控制设备基于各个车道线网格的语义信息对局部地图中的各个车道线网格进行连通域分析处理时，可以首先根据各个车道线网格的语义信息从局部地图中确定出各个车道线网格，并对各个车道线网格进行连通域提取，根据各个车道线网格的语义信息确定小范围内车道线网格的连通域标签，基于各车道线网格对应的连通域标签可进行车道线拟合，确定出小范围内车道线位置点对应的一条或多条局部车道线，以此类推，可确定出局部地图中的所有局部车道线，从而得到局部地图对应的局部车道线集合。
其中，属于同一车道线的车道线网格对应相同的连通域标签，该连通域标签与车道线网格的语义信息关联。例如，车道线网格G1的语义信息用于指示该车道线网格G1对应的图像点属于车道线A，车道线网格G2的语义信息用于指示该车道线网格G2对应的图像点也属于车道线A，那么该车道线网格G1和车道线网格G2对应相同的连通域标签。
如图5所示，为局部地图中添加的连通域标签的示意图，如图所示，每一个网格表示一个图像点，网格中的数字表示各网格对应的连通域标签，其中，连通域标签为0的网格可能为局部车道线中不属于车道线的网格，连通域标签 不为0的为可能属于车道线的车道线网格。连通域标签为1的网格为局部地图中属于A车道线的图像点；连通域标签为2的图像点为局部地图中属于B车道线对应的图像点，其中，该A车道线和该B车道线为不相同的两条车道线。
在一个实施例中，控制设备在对局部地图中进行连通域分析处理时，可按照预设尺寸的图像检测窗口从局部地图中选取小范围的网格，并基于各个网格的语义信息确定出处于该图像检测窗口内的各网格的连通域标签，其中，非车道线网格对应的连通域标签为0，车道线网格的连通域标签由对应的语义信息确定。由于处于相同连通域的网格对应有相同的连通域标签，所以基于连通域标签是否相同的特性可确定出图像检测窗口的各连通域，即确定每个连通域包括的网格，其中，所述图像检测窗口的预设尺寸例如可以是3*3或者5*5。
在确定每个连通域包括的各网格后，可对各连通域对应的网格按照最优解算法进行拟合，得到各连通域对应的局部车道线，以此类推，可确定出图像检测窗口在局部地图中位于不同位置时包括的车道线对应的局部车道线，从而可确定出局部地图中的所有局部车道线，以得到局部车道线集合。
举例来说，如果从局部车道线中选取的位于图像检测窗口中的网格对应的连通域标签如图5所示，可将连通域标签为1的所有车道线网格按照最优解算法拟合出一条局部车道线，如图中的501，将连通域标签为2的所有车道线网格按照最优解算法拟合出一条局部车道线，如图中的502，以此类推，可将处于该图像检测窗口中的所有车道线网格拟合成多条局部车道线，从而可确定出局部地图中的所有局部车道线。
进一步地，在控制设备确定出局部地图对应的局部车道线集合之后，可以基于带权最大团算法和局部车道线集合中各局部车道线的权重值，优化确定出至少一条初始车道线，并对至少一条初始车道线进行过滤处理，得到初始车道线集合。
其中，局部车道线的权重值是根据车道线的特征信息确定的，在一个实施例中，车道线的特征信息包括：车道线的几何特征和/或颜色特征，几何特征包括长度特征、宽度特征以及车道线之间的平行特征中的任意一种或多种。
作为一种可行的实施例方式，可以预先建立车道线的几何特征、颜色特征与假设分值的对应关系，一个车道线对应的假设分值总和越高，那么该车道线 为实际车道线的可能性越大。示例性地，该车道线的几何特征、颜色特征与假设分值的对应关系可以如表1所示。可以看出，每一个局部车道线均可以通过该表1所示的对应关系，确定出自身对应的假设分值总和，该假设分值总和即为该局部车道线对应的权重值。
表1
Figure PCTCN2019084934-appb-000001
其中，表1中各个特征维度对应的假设分值仅为一种示例，主要用于说明通过车道线的特征信息确定局部车道线权重值的原理，不能成为对本发明实施例的限定。
在一个实施例中，控制设备可以解析各个局部车道线的特征信息，并根据预先建立的车道线的几何特征、颜色特征与假设分值的对应关系，确定出各个局部车道线的权重值。进一步地，基于带权最大团算法，解出局部车道线中权重值最高的至少一条局部车道线，即至少一条初始车道线。
进一步地，控制设备可以对至少一条初始车道线进行过滤处理，得到初始车道线集合。在一个实施例中，可基于预设的针对初始车道线的先验信息进行后处理，以实现对上述至少一条初始车道线的所有初始车道线进行过滤，将该所有初始车道线中的错误车道线滤除，得到初始车道线集合。
其中，所述预设的针对初始车道线的先验信息是基于车道线的国家标准设定的一个预设范围或者预设值，所述先验信息具体包括长度信息以及宽度信息等，假设国家标准的车道线长度为1.5米，则该针对初始车道线的先验信息包 括的长度信息可以设定为小于标准车道线长度的部分长度，例如10厘米～15厘米的范围等等；假设国家标准的车道线宽度为15厘米，则该先验信息包括的宽度信息可以设定为13厘米～17厘米的范围等。对应的，该错误车道线为不满足该预设的先验信息的车道线。
控制设备在从局部地图中确定出初始车道线集合之后，控制设备可以在S303中对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。可以将当前的车道线检测结果与历史车道线检测结果进行时序和空间上的融合，有利于提高车道线检测结果的准确度。
在一个实施例中，控制设备在从局部地图中确定出初始车道线集合之后，可以查询存储区域中是否存在历史车道线集合，若不存在该历史车道线集合，则可以确定本次车道线检测为首次车道线检测，并将本次获得的初始车道线集合作为历史车道线集合存储在存储区域中，以便于后续与新的初始车道线集合进行拟合优化。若存在历史车道线集合，则可以对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
相对于直接根据环境图像中所有车道线图像点拟合得到完整车道线的方法，采用本发明实施例的从可移动平台所处环境的局部地图出发，构造整体车道线的方法，一方面可在一定程度上提高车道线的拟合精度，提高车道线识别的准确度，另一方面，由于局部地图所占区域比较小，还可在一定程度上节省可移动平台在进行车道线拟合时运算速度。
本发明实施例中，控制设备可以获取可移动平台的环境图像，并根据环境图像得到可移动平台所处环境的局部地图，从局部地图中确定出初始车道线集合，并对历史车道线集合中的历史车道线数据和初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。可以不依赖于车道线几何平行假设和路面平行假设，实现对车道线的检测，有利于更好地适用于多种车道线场景，提高通用性。
请参见图6，是本发明实施例提出的另一种车道线检测方法的流程示意图，该方法应用于控制设备，该控制设备与可移动平台之间存在数据连接，其中， 所述方法包括以下步骤：
S601，获取可移动平台的环境图像，并根据环境图像得到可移动平台所处环境的局部地图，其中，在局部地图中记录了车道线位置点。进一步地，控制设备可以在S602中从局部地图中确定出初始车道线集合，该初始车道线集合包括多个初始车道线数据。进一步地，控制设备在从局部地图中确定出初始车道线集合之后，可以在S603中从初始车道线集合中确定出目标初始车道线，并从历史车道线集合中确定出与目标初始车道线匹配的目标历史车道线，该目标初始车道线为所述初始车道线集合中的任一个。
其中，目标初始车道线与目标历史车道线对应的索引信息相匹配。在一个实施例中，局部地图可以为在世界坐标系下的网格地图，该车道线位置点对应的网格为车道线网格，每个车道线网格包括一个语义信息，该语义信息用于表征车道线网格为车道线位置点对应的网格。针对这种情况下，控制设备从局部地图中确定出初始车道线集合之后，由于初始车道线集合中的各个初始车道线由至少一个车道线网格进行曲线拟合而成，因此，可以基于组成初始车道线的车道线网格的语音信息确定各个初始车道线所属的车道，进而对各个初始车道线添加索引信息，每个索引信息指示了初始车道线所属的车道。
进一步地，可以从初始车道线集合中选取任一个初始车道线作为目标车道线，并将该目标车道线的索引信息与历史车道线集合中各个历史车道线对应的索引信息进行对比，若对比得到该目标车道线的索引信息与历史车道线集合中任一历史车道线对应的索引信息相同，则可以确定该目标车道线的索引信息与历史车道线集合中任一历史车道线对应的索引信息匹配，并将该任一历史车道线确定为目标历史车道线。
进一步地，依次类推，控制设备可以从初始车道线集合中选取下一个目标车道线，并再次检测历史车道线集合中是否存在与该下一个目标车道线匹配的目标历史车道线，依次类推，直到将初始车道线集合中的所有初始车道线检测完毕，则结束。
示例性地，参见图7，假设控制设备从局部地图中确定出的初始车道线集合如图中的60所示，该初始车道线集合60包括初始车道线1和初始车道线2，初始车道线1的索引信息指示该初始车道线1属于车道线A，初始车道线2的 索引信息指示该初始车道线2属于车道线B；历史车道线集合如图中的61所示，该历史车道线集合包括历史车道线1和历史车道线2，历史车道线1的索引信息指示该历史车道线1属于车道线A，历史车道线2的索引信息指示该历史车道线2属于车道线B。针对这种情况下，控制设备可以将初始车道线1和初始车道线2各自对应的索引信息，与历史车道线1和历史车道线2各自对应的索引信息进行对比，由于初始车道线1和历史车道线1的索引信息均指示属于车道线A，可以确定该初始车道线1与历史车道线1匹配，也即历史车道线1为初始车道线1对应的目标历史车道线；由于初始车道线2和历史车道线2的索引信息均指示属于车道线B，可以确定该初始车道线2与历史车道线2匹配，也即历史车道线2为初始车道线2对应的目标历史车道线。
在控制设备从历史车道线集合中确定出与目标初始车道线匹配的目标历史车道线之后，可以在步骤S604中根据过拟合约束条件和平行约束条件，对目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线，并由至少一个目标车道线组成目标车道线集合。
在一个实施例中，可以根据过拟合约束条件和平行约束条件创建的车道模型对目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线。其中，每条车道线可以采用独立的车道模型，该车道模型主要用于对目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线。其中，车道模型生成目标车道线的过程，其实质就是一个不断压缩新的目标初始车道线到目标历史车道线的过程，从而实现目标初始车道线与目标历史车道线的融合。
示例性地，该车道模型对应的数学函数可以由三部分组成，第一部分为基础的最小二乘模型，用于求解目标初始车道线对应的车道线观测点云与目标历史车道线为对应的曲线方程之间的最小误差；第二部分为曲线的平滑项(对应过拟合预设条件)，可以使用对曲线方程的三阶导数平方进行积分作为平滑约束，主要用于防止过拟合；第三部分为弱平行约束项(对应平行约束条件)，主要用于对相邻两车道线间的平行状态进行约束。
进一步地，参见图7，假设控制设备从局部地图中确定出的初始车道线集合如图中的60所示，历史车道线集合如图中的61所示，控制设备确定出历史车道线1为初始车道线1对应的目标历史车道线，历史车道线2为初始车道线2对应的目标历史车道线。这种情况下，控制设备可以基于该车道模型将初始车道线1和历史车道线1对齐，融合初始车道线1和历史车道线1，得到目标车道线621；基于该车道模型将初始车道线2和历史车道线2对齐，融合初始车道线2和历史车道线2，得到目标车道线622。进一步地，所有的目标车道线(即目标车道线621和目标车道线622)组成目标车道线集合62。
其中，由于每条车道线采用独立的车道模型进行目标初始车道线和目标历史车道线的拟合，且在相邻两车道线间增加了平行状态的约束。因此，对于复杂的车道检测场景(如分岔路场景、城区场景等)，均可以保证车道线检测的准确性，从而能够更好地适用于多种车道线场景，提升了通用性。
在一个实施例中，控制设备得到目标车道线集合之后，由于得到的目标车道线均为车道边界线，进一步地，控制设备可以基于车道的属性信息对目标车道线集合中的目标车道线进行组合，得到至少一个车道，并生成车道的车道中心线，以便于辅助可移动平台驾驶。该属性信息包括车道的几何特征和/或颜色特征，该几何特征包括长度特征、宽度特征以及车道之间的平行特征中的任意一种或多种。
在本发明实施例中，控制设备可以获取可移动平台的环境图像，并根据环境图像得到所述可移动平台所处环境的局部地图，从局部地图中确定出初始车道线集合。进一步地，从初始车道线集合中确定出目标初始车道线，并从历史车道线集合中确定出与目标初始车道线匹配的目标历史车道线，进而根据过拟合约束条件和平行约束条件，对目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线。通过增加相邻两车道间的平行约束条件，有利于更好地适用于多种车道线场景，提升通用性。
本发明实施例提供了一种车道线检测装置，所述车道线检测装置用于执行前述任一项所述的方法的模块，具体地，参见图8，是本发明实施例提供的一 种车道线检测装置的示意框图，本实施例的车道线检测装置可配置于控制设备，该控制设备可以与例如自动驾驶汽车等类型的可移动平台存在数据连接，车道线检测装置包括：获取模块80和处理模块81。
其中，获取模块80，用于获取所述可移动平台的环境图像。
处理模块81，用于根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
在一个实施例中，所述车道线位置点是通过对所述环境图像进行图像分析确定的。
在一个实施例中，所述初始车道线集合中包括的初始车道线数据是根据所述局部地图中记录的车道线位置点确定的。
在一个实施例中，处理模块81，具体用于基于图像模型识别所述环境图像中的车道线图像点；
基于所述可移动平台的位姿信息和所述环境图像得到所处环境的局部地图；
根据所述车道线图像点确定出所述局部地图中的车道线位置点。
在一个实施例中，所述局部地图为在世界坐标系下的网格地图，所述车道线位置点对应的网格为车道线网格，每个车道线网格包括一个语义信息，所述语义信息用于表征所述车道线网格为所述车道线位置点对应的网格。
在一个实施例中，处理模块81，具体用于对所述局部地图中的各个车道线网格进行分析处理，得到所述局部地图对应的局部车道线集合，所述局部车道线集合包括至少一个局部车道线；从所述局部车道线集合中确定出初始车道线集合。
在一个实施例中，处理模块81，具体用于基于各个车道线网格的语义信息对所述局部地图中的各个车道线网格进行连通域分析处理，得到所述局部地图对应的局部车道线集合。
在一个实施例中，处理模块81，具体用于基于预设的优化算法和所述局 部车道线集合中各局部车道线的权重值，优化确定出初始车道线，得到初始车道线集合；其中，所述局部车道线的权重值是根据车道线的特征信息确定的。
在一个实施例中，处理模块81，具体用于基于带权最大团算法和所述局部车道线集合中各局部车道线的权重值，优化确定出至少一条初始车道线；对所述至少一条初始车道线进行过滤处理，得到初始车道线集合。
在一个实施例中，所述车道线的特征信息包括：所述车道线的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道线之间的平行特征中的任意一种或多种。
在一个实施例中，处理模块81，具体用于从所述初始车道线集合中确定出目标初始车道线，并从历史车道线集合中确定出与所述目标初始车道线匹配的目标历史车道线，所述目标初始车道线为所述初始车道线集合中的任一个；
根据过拟合约束条件和平行约束条件，对所述目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线；由至少一个所述目标车道线组成目标车道线集合。
在一个实施例中，所述目标初始车道线与目标历史车道线对应的索引信息相匹配。
在一个实施例中，处理模块81，还用于基于车道的属性信息对所述目标车道线集合中的目标车道线进行组合，得到至少一个车道，并生成所述车道的车道中心线；所述属性信息包括所述车道的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道之间的平行特征中的任意一种或多种。
在一个实施例中，本实施例提供的车道线检测装置能执行前述实施例提供的如图3和图6所示的车道线检测方法，且执行方式和有益效果类似，在这里不再赘述。
本发明实施例提供了一种控制设备，所述控制设备可配置于上述实施例提及的可移动平台，其中，图9是本发明实施例提供的一种控制设备的结构图，如图9所示，所述控制设备90包括存储器901、处理器902和视觉传感器903。
其中，所述处理器902可以是中央处理器(central processing unit，CPU)。所述处理器902可以是硬件芯片。所述硬件芯片可以是专用集成电路(application-specific integrated circuit，ASIC)，可编程逻辑器件(programmable  logic device，PLD)或其组合。所述PLD可以是复杂可编程逻辑器件(complex programmable logic device，CPLD)，现场可编程逻辑门阵列(field-programmable gate array，FPGA)，通用阵列逻辑(generic array logic，GAL)或其任意组合。
存储器901中存储有程序代码，处理器902调用存储器中的程序代码，当程序代码被执行时，处理器902执行如下操作：通过视觉传感器903获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
在一个实施例中，所述车道线位置点是通过对所述环境图像进行图像分析确定的。
在一个实施例中，所述初始车道线集合中包括的初始车道线数据是根据所述局部地图中记录的车道线位置点确定的。
在一个实施例中，处理器902，具体用于基于图像模型识别所述环境图像中的车道线图像点；
基于所述可移动平台的位姿信息和所述环境图像得到所处环境的局部地图；
根据所述车道线图像点确定出所述局部地图中的车道线位置点。
在一个实施例中，所述局部地图为在世界坐标系下的网格地图，所述车道线位置点对应的网格为车道线网格，每个车道线网格包括一个语义信息，所述语义信息用于表征所述车道线网格为所述车道线位置点对应的网格。
在一个实施例中，处理器902，具体用于对所述局部地图中的各个车道线网格进行分析处理，得到所述局部地图对应的局部车道线集合，所述局部车道线集合包括至少一个局部车道线；从所述局部车道线集合中确定出初始车道线集合。
在一个实施例中，处理器902，具体用于基于各个车道线网格的语义信息对所述局部地图中的各个车道线网格进行连通域分析处理，得到所述局部地图 对应的局部车道线集合。
在一个实施例中，处理器902，具体用于基于预设的优化算法和所述局部车道线集合中各局部车道线的权重值，优化确定出初始车道线，得到初始车道线集合；其中，所述局部车道线的权重值是根据车道线的特征信息确定的。
在一个实施例中，处理器902，具体用于基于带权最大团算法和所述局部车道线集合中各局部车道线的权重值，优化确定出至少一条初始车道线；对所述至少一条初始车道线进行过滤处理，得到初始车道线集合。
在一个实施例中，所述车道线的特征信息包括：所述车道线的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道线之间的平行特征中的任意一种或多种。
在一个实施例中，处理器902，具体用于从所述初始车道线集合中确定出目标初始车道线，并从历史车道线集合中确定出与所述目标初始车道线匹配的目标历史车道线，所述目标初始车道线为所述初始车道线集合中的任一个；
根据过拟合约束条件和平行约束条件，对所述目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线；由至少一个所述目标车道线组成目标车道线集合。
在一个实施例中，所述目标初始车道线与目标历史车道线对应的索引信息相匹配。
在一个实施例中，处理器902，还用于基于车道的属性信息对所述目标车道线集合中的目标车道线进行组合，得到至少一个车道，并生成所述车道的车道中心线；所述属性信息包括所述车道的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道之间的平行特征中的任意一种或多种。
本实施例提供的可移动平台能执行前述实施例提供的如图3和图6所示的车道线检测方法，且执行方式和有益效果类似，在这里不再赘述。
本发明实施例还提供一种包含指令的计算机程序产品，当其在计算机上运行时，使得计算机执行上述方法实施例所述的车道线检测方法的相关步骤。
本领域普通技术人员可以理解实现上述实施例方法中的全部或部分流程，是可以通过计算机程序来指令相关的硬件来完成，所述的程序可存储于一计算机可读取存储介质中，该程序在执行时，可包括如上述各方法的实施例的流程。 其中，所述的存储介质可为磁碟、光盘、只读存储记忆体(Read-Only Memory，ROM)或随机存储记忆体(Random Access Memory，RAM)等。
以上所揭露的仅为本发明部分实施例而已，当然不能以此来限定本发明之权利范围，因此依本发明权利要求所作的等同变化，仍属本发明所涵盖的范围。

Claims (28)

  1. 一种车道线检测方法，其特征在于，所述方法应用于控制设备，所述控制设备与可移动平台之间存在数据连接，该方法包括：
    获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
    从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
    对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
  2. 根据权利要求1所述的方法，其特征在于，所述车道线位置点是通过对所述环境图像进行图像分析确定的。
  3. 根据权利要求1所述的方法，其特征在于，所述初始车道线集合中包括的初始车道线数据是根据所述局部地图中记录的车道线位置点确定的。
  4. 根据权利要求1所述的方法，其特征在于，所述根据所述环境图像得到所述可移动平台所处环境的局部地图，包括：
    基于图像模型识别所述环境图像中的车道线图像点；
    基于所述可移动平台的位姿信息和所述环境图像得到所处环境的局部地图；
    根据所述车道线图像点确定出所述局部地图中的车道线位置点。
  5. 根据权利要求1-4任一项所述的方法，其特征在于，所述局部地图为在世界坐标系下的网格地图，所述车道线位置点对应的网格为车道线网格，每个车道线网格包括一个语义信息，所述语义信息用于表征所述车道线网格为所述车道线位置点对应的网格。
  6. 根据权利要求5所述的方法，其特征在于，所述从所述局部地图中确定出初始车道线集合，包括：
    对所述局部地图中的各个车道线网格进行分析处理，得到所述局部地图对应的局部车道线集合，所述局部车道线集合包括至少一个局部车道线；
    从所述局部车道线集合中确定出初始车道线集合。
  7. 根据权利要求6所述的方法，其特征在于，所述对所述局部地图中的各个车道线网格进行分析处理，得到所述局部地图对应的局部车道线集合，包括：
    基于各个车道线网格的语义信息对所述局部地图中的各个车道线网格进行连通域分析处理，得到所述局部地图对应的局部车道线集合。
  8. 根据权利要求6所述的方法，其特征在于，所述从所述局部车道线集合中确定出初始车道线集合，包括：
    基于预设的优化算法和所述局部车道线集合中各局部车道线的权重值，优化确定出初始车道线，得到初始车道线集合；
    其中，所述局部车道线的权重值是根据车道线的特征信息确定的。
  9. 根据权利要求8所述的方法，其特征在于，所述基于预设的优化算法和所述局部车道线集合中各局部车道线的权重值，优化确定出初始车道线，得到初始车道线集合，包括：
    基于带权最大团算法和所述局部车道线集合中各局部车道线的权重值，优化确定出至少一条初始车道线；
    对所述至少一条初始车道线进行过滤处理，得到初始车道线集合。
  10. 根据权利要求8所述的方法，其特征在于，所述车道线的特征信息包括：所述车道线的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道线之间的平行特征中的任意一种或多种。
  11. 根据权利要求1-10任一项所述的方法，其特征在于，所述对历史车道线集合中历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合，包括：
    从所述初始车道线集合中确定出目标初始车道线，并从历史车道线集合中确定出与所述目标初始车道线匹配的目标历史车道线，所述目标初始车道线为所述初始车道线集合中的任一个；
    根据过拟合约束条件和平行约束条件，对所述目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线；
    由至少一个所述目标车道线组成目标车道线集合。
  12. 根据权利要求11所述的方法，其特征在于，所述目标初始车道线与目标历史车道线对应的索引信息相匹配。
  13. 根据权利要求1-12任一项所述的方法，其特征在于，所述得到目标车道线集合之后，所述方法还包括：
    基于车道的属性信息对所述目标车道线集合中的目标车道线进行组合，得到至少一个车道，并生成所述车道的车道中心线；
    所述属性信息包括所述车道的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道之间的平行特征中的任意一种或多种。
  14. 一种车道线检测装置，其特征在于，所述装置配置于控制设备，该控制设备与可移动平台之间存在数据连接，该装置包括：
    获取模块，用于获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
    处理模块，用于从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
    所述处理模块，还用于对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
  15. 一种控制设备，其特征在于，所述控制设备与可移动平台之间存在数据连接，该控制设备包括：视觉传感器和处理器，所述处理器用于：
    通过所述视觉传感器获取所述可移动平台的环境图像，并根据所述环境图像得到所述可移动平台所处环境的局部地图，其中，在所述局部地图中记录了车道线位置点；
    从所述局部地图中确定出初始车道线集合，所述初始车道线集合包括多个初始车道线数据；
    对历史车道线集合中的历史车道线数据和所述初始车道线集合中的初始车道线数据进行拟合优化，得到目标车道线集合。
  16. 根据权利要求15所述的控制设备，其特征在于，所述车道线位置点是通过对所述环境图像进行图像分析确定的。
  17. 根据权利要求15所述的控制设备，其特征在于，所述初始车道线集合中包括的初始车道线数据是根据所述局部地图中记录的车道线位置点确定的。
  18. 根据权利要求15所述的控制设备，其特征在于，所述处理器，具体用于基于图像模型识别所述环境图像中的车道线图像点；基于所述可移动平台的位姿信息和所述环境图像得到所处环境的局部地图；根据所述车道线图像点确定出所述局部地图中的车道线位置点。
  19. 根据权利要求15-18任一项所述的控制设备，其特征在于，所述局部地图为在世界坐标系下的网格地图，所述车道线位置点对应的网格为车道线网格，每个车道线网格包括一个语义信息，所述语义信息用于表征所述车道线网 格为所述车道线位置点对应的网格。
  20. 根据权利要求19所述的控制设备，其特征在于，所述处理器，具体用于对所述局部地图中的各个车道线网格进行分析处理，得到所述局部地图对应的局部车道线集合，所述局部车道线集合包括至少一个局部车道线；从所述局部车道线集合中确定出初始车道线集合。
  21. 根据权利要求20所述的控制设备，其特征在于，所述处理器，具体用于基于各个车道线网格的语义信息对所述局部地图中的各个车道线网格进行连通域分析处理，得到所述局部地图对应的局部车道线集合。
  22. 根据权利要求20所述的控制设备，其特征在于，所述处理器，具体用于基于预设的优化算法和所述局部车道线集合中各局部车道线的权重值，优化确定出初始车道线，得到初始车道线集合；其中，所述局部车道线的权重值是根据车道线的特征信息确定的。
  23. 根据权利要求22所述的控制设备，其特征在于，所述处理器，具体用于基于带权最大团算法和所述局部车道线集合中各局部车道线的权重值，优化确定出至少一条初始车道线；对所述至少一条初始车道线进行过滤处理，得到初始车道线集合。
  24. 根据权利要求22所述的控制设备，其特征在于，所述车道线的特征信息包括：所述车道线的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道线之间的平行特征中的任意一种或多种。
  25. 根据权利要求15-24任一项所述的控制设备，其特征在于，所述处理器，还具体用于从所述初始车道线集合中确定出目标初始车道线，并从历史车道线集合中确定出与所述目标初始车道线匹配的目标历史车道线，所述目标初始车道线为所述初始车道线集合中的任一个；根据过拟合约束条件和平行约束 条件，对所述目标初始车道线对应的初始车道线数据和目标历史车道线对应的历史车道线数据进行拟合优化，得到目标车道线；由至少一个所述目标车道线组成目标车道线集合。
  26. 根据权利要求25所述的控制设备，其特征在于，所述目标初始车道线与目标历史车道线对应的索引信息相匹配。
  27. 根据权利要求15-26任一项所述的控制设备，其特征在于，所述处理器，还用于基于车道的属性信息对所述目标车道线集合中的目标车道线进行组合，得到至少一个车道，并生成所述车道的车道中心线；所述属性信息包括所述车道的几何特征和/或颜色特征，所述几何特征包括长度特征、宽度特征以及车道之间的平行特征中的任意一种或多种。
  28. 一种计算机可读存储介质，其特征在于，所述计算机可读存储介质中存储有程序指令，当所述程序指令在处理器上运行时，实现权利要求1-13任一项所述的车道线检测方法。
PCT/CN2019/084934 2019-04-29 2019-04-29 一种车道线检测方法、装置、控制设备及存储介质 WO2020220182A1 (zh)

Priority Applications (2)

Application Number Priority Date Filing Date Title
CN201980004970.XA CN111295666A (zh) 2019-04-29 2019-04-29 一种车道线检测方法、装置、控制设备及存储介质
PCT/CN2019/084934 WO2020220182A1 (zh) 2019-04-29 2019-04-29 一种车道线检测方法、装置、控制设备及存储介质

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2019/084934 WO2020220182A1 (zh) 2019-04-29 2019-04-29 一种车道线检测方法、装置、控制设备及存储介质

Publications (1)

Publication Number Publication Date
WO2020220182A1 true WO2020220182A1 (zh) 2020-11-05

Family

ID=71027421

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2019/084934 WO2020220182A1 (zh) 2019-04-29 2019-04-29 一种车道线检测方法、装置、控制设备及存储介质

Country Status (2)

Country Link
CN (1) CN111295666A (zh)
WO (1) WO2020220182A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220196408A1 (en) * 2019-09-09 2022-06-23 Huawei Technologies Co., Ltd. Lane Line Information Determining Method and Apparatus
CN115116017A (zh) * 2022-06-27 2022-09-27 北京旋极信息技术股份有限公司 车道识别方法、装置及存储介质

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112215144B (zh) * 2020-10-12 2024-05-14 北京四维智联科技有限公司 一种处理车道线的方法及系统
CN114359871A (zh) * 2021-12-31 2022-04-15 中国第一汽车股份有限公司 车辆可行驶区域的区域边缘检测方法、装置和存储介质
CN114691659A (zh) * 2022-03-30 2022-07-01 阿波罗智联(北京)科技有限公司 一种地图数据的优化方法、装置、设备以及存储介质
CN115240435A (zh) * 2022-09-21 2022-10-25 广州市德赛西威智慧交通技术有限公司 一种基于ai技术的车辆违章行驶检测方法、装置
CN116499477B (zh) * 2023-06-21 2023-09-26 小米汽车科技有限公司 地图的融合方法、装置、介质及车辆

Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6819779B1 (en) * 2000-11-22 2004-11-16 Cognex Corporation Lane detection system and apparatus
CN107330376A (zh) * 2017-06-06 2017-11-07 广州汽车集团股份有限公司 一种车道线识别方法及系统
CN109426800A (zh) * 2017-08-22 2019-03-05 北京图森未来科技有限公司 一种车道线检测方法和装置

Family Cites Families (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US8311283B2 (en) * 2008-07-06 2012-11-13 Automotive Research&Testing Center Method for detecting lane departure and apparatus thereof

Patent Citations (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6819779B1 (en) * 2000-11-22 2004-11-16 Cognex Corporation Lane detection system and apparatus
CN107330376A (zh) * 2017-06-06 2017-11-07 广州汽车集团股份有限公司 一种车道线识别方法及系统
CN109426800A (zh) * 2017-08-22 2019-03-05 北京图森未来科技有限公司 一种车道线检测方法和装置

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20220196408A1 (en) * 2019-09-09 2022-06-23 Huawei Technologies Co., Ltd. Lane Line Information Determining Method and Apparatus
CN115116017A (zh) * 2022-06-27 2022-09-27 北京旋极信息技术股份有限公司 车道识别方法、装置及存储介质

Also Published As

Publication number Publication date
CN111295666A (zh) 2020-06-16

Similar Documents

Publication Publication Date Title
WO2020220182A1 (zh) 一种车道线检测方法、装置、控制设备及存储介质
US20200302189A1 (en) Lane line data processing method and apparatus, computer device, and storage medium
Chen et al. Pointlanenet: Efficient end-to-end cnns for accurate real-time lane detection
US11783594B2 (en) Method of segmenting pedestrians in roadside image by using convolutional network fusing features at different scales
US11367217B2 (en) Image processing method and apparatus, and related device
JP7120689B2 (ja) 深層学習に基づく車載ビデオターゲット検出方法
CN111444821A (zh) 一种城市道路标志自动识别方法
CN103500322B (zh) 基于低空航拍图像的车道线自动识别方法
WO2021056341A1 (zh) 车道线融合方法、车道线融合装置、车辆和存储介质
WO2022134996A1 (en) Lane line detection method based on deep learning, and apparatus
WO2020181426A1 (zh) 一种车道线检测方法、设备、移动平台及存储介质
CN111046767B (zh) 一种基于单目图像的3d目标检测方法
Mei et al. Scene-adaptive off-road detection using a monocular camera
CN113240623B (zh) 一种路面病害检测方法及装置
KR20200102378A (ko) 정보 처리 방법, 장치 및 저장 매체
JP7119197B2 (ja) 車線属性検出
Jiang et al. Deep transfer learning enable end-to-end steering angles prediction for self-driving car
CN112037180B (zh) 染色体分割方法及装置
CN111222522A (zh) 神经网络训练、路面检测、智能驾驶控制方法和装置
CN104915642A (zh) 前方车辆测距方法及装置
CN111098850A (zh) 一种自动停车辅助系统及自动泊车方法
WO2024183419A1 (zh) 可行驶区域检测方法、系统及计算机可读介质
CN109543498B (zh) 一种基于多任务网络的车道线检测方法
CN110909656B (zh) 一种雷达与摄像机融合的行人检测方法和系统
CN111428616B (zh) 一种车位检测方法、装置、设备及存储介质

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 19926817

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 19926817

Country of ref document: EP

Kind code of ref document: A1