WO2024002014A1 - 交通标线的识别方法、装置、计算机设备和存储介质 - Google Patents

交通标线的识别方法、装置、计算机设备和存储介质 Download PDF

Info

Publication number
WO2024002014A1
WO2024002014A1 PCT/CN2023/102444 CN2023102444W WO2024002014A1 WO 2024002014 A1 WO2024002014 A1 WO 2024002014A1 CN 2023102444 W CN2023102444 W CN 2023102444W WO 2024002014 A1 WO2024002014 A1 WO 2024002014A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
probability
sample
point cloud
grid
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.)
Ceased
Application number
PCT/CN2023/102444
Other languages
English (en)
French (fr)
Inventor
宋潇
王哲
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shanghai Sensetime Intelligent Technology Co Ltd
Original Assignee
Shanghai Sensetime Intelligent Technology Co Ltd
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 Shanghai Sensetime Intelligent Technology Co Ltd filed Critical Shanghai Sensetime Intelligent Technology Co Ltd
Publication of WO2024002014A1 publication Critical patent/WO2024002014A1/zh
Anticipated expiration legal-status Critical
Ceased legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; 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 OR CALCULATING; COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
    • G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
    • G—PHYSICS
    • G06—COMPUTING OR CALCULATING; COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks

Definitions

  • the present disclosure relates to the field of computer vision technology, and specifically, to a traffic marking recognition method, device, computer equipment and storage medium.
  • road images are usually captured by a camera device installed on the vehicle, and then based on the road image, the distance between traffic markings such as lane lines and road boundary lines on the road and the vehicle are determined.
  • lane lines and road boundary lines determined based on road images usually have discontinuous and inaccurate line segments.
  • Embodiments of the present disclosure provide at least a method, device, computer equipment, and storage medium for identifying traffic markings.
  • embodiments of the present disclosure provide a method for identifying traffic markings, including:
  • the marking recognition result of the road point cloud data is determined.
  • the road point cloud data captured by lidar has morphological stability
  • using the acquired road point cloud data to identify traffic marking categories can improve the continuity of the identified traffic markings.
  • the probability that each target grid belongs to the traffic marking category is determined. Based on the probability of the traffic marking category to which each target grid belongs, each target grid belonging to the traffic marking category can be accurately determined from multiple target grids. grid, and then use each target raster belonging to the traffic marking category to accurately determine each point cloud point belonging to the traffic marking category in the road point cloud data, and then accurately obtain the traffic markings of each category, that is, the marking recognition result .
  • the traffic marking category includes a lane line category and a road boundary line category; the target grid is determined based on the local point cloud data contained in each target grid.
  • the probability that the grid belongs to the traffic marking category includes:
  • a first probability that the target grid belongs to the lane line category and a second probability that the target grid belongs to the road boundary line category are determined.
  • the feature information related to the traffic marking category in each point cloud point can be fully extracted, and then the accurate image of the target grid can be obtained.
  • Road feature information Using the road feature information to determine the first probability that the target raster belongs to the lane line category and the second probability that it belongs to the road boundary line category, the target raster can be accurately classified into the corresponding traffic marking category, and then the target raster can be obtained. Accurate traffic marking category.
  • the traffic marking category based on each of the target grids respectively belongs to Probability, determine the marking recognition result of the road point cloud data, including:
  • a marking line identification result of the road point cloud data is determined.
  • the preset threshold can be used to filter out target probabilities with larger probability values.
  • the traffic marking category associated with the target probability is more likely to correspond to the traffic marking category to which the target grid belongs. Therefore, based on the target The probabilistically associated traffic marking categories determine the recognition results more accurately. In this way, by using each target probability to determine the recognition results of each target grid, and then determining the marking line recognition results based on each recognition result, the accuracy of the determined marking line recognition results can be improved.
  • determining the recognition result of the target grid based on the traffic marking category associated with the target probability includes:
  • the target probability includes the first probability and the second probability, determining the maximum probability among the first probability and the second probability;
  • the traffic marking category associated with the maximum probability is used as the identification result of the target grid.
  • the traffic marking category associated with the maximum probability is used as the identification result of the target grid, which can ensure the uniqueness of the obtained target probability and improve the obtained Identify the reasonableness of the results.
  • the road point cloud data is collected by a driving device, and after determining the marking line recognition result of the road point cloud data, the method further includes:
  • the traveling device by controlling the driving of the traveling device through the determined marking position and/or marking type, it can ensure that the traveling device travels in a reasonable area and improve the driving safety of the traveling device.
  • determining the probability that the target grid belongs to the traffic marking category based on the local point cloud data contained in each target grid includes:
  • the probability that the target grid belongs to the traffic marking category is determined based on the local point cloud data contained in each target grid.
  • the trained target neural network has reliable prediction accuracy. Using the trained target neural network, the probability that the target grid belongs to the traffic markings of each category can be accurately determined.
  • the target neural network is trained according to the following steps:
  • the neural network to be trained is iteratively trained until the training cutoff condition is met, and the result is obtained Describe the target neural network.
  • the predicted probability includes a first predicted probability that the sample raster belongs to the lane line category, and a second predicted probability that the sample raster belongs to the road boundary line category;
  • the neural network to be trained is iteratively trained until the training cutoff condition is met, Obtain the target neural network, including:
  • the annotation label information of the sample grid indicates that the sample grid does not belong to the background, based on The first predicted probability, the second predicted probability of the sample grid, the label information of the sample grid, and the label information corresponding to the background label determine the first loss;
  • the label information of the sample grid indicates that the sample grid belongs to the background, based on the first predicted probability, the second predicted probability and the label information corresponding to the background label, it is determined second loss;
  • the neural network to be trained is iteratively trained until a training cutoff condition is met, and the target neural network is obtained.
  • the consistency between the prediction probability of the network output and the annotation label information can be improved, thereby obtaining the target neural network with reliable prediction accuracy. network.
  • the first loss includes a first sub-loss and a second sub-loss
  • Determining the first loss based on the first prediction probability, the second prediction probability of the sample grid, the label information of the sample grid, and the label information corresponding to the background label includes:
  • the second sub-loss is determined based on other prediction probabilities in the first prediction probability and the second prediction probability except the target prediction probability, and annotation label information corresponding to the background label.
  • the first sub-loss can represent the difference between the predicted probability output by the network and the annotation label information
  • the second sub-loss can represent the difference between the predicted probability output by the network and the annotation label information corresponding to the background label.
  • the first sub-loss and the second sub-loss train the network, which can improve the consistency between the prediction probability of the network output and the annotation label information.
  • determining the annotation label information of each sample grid includes:
  • a top view of the sample is generated; wherein each sample grid corresponds to a pixel in the top view of the sample;
  • the annotation label information of the sample grid matching each pixel is determined.
  • a top view of the sample is generated based on the local sample point cloud data contained in each sample grid, so that the pixel information of the pixels in the top view of the sample can be used to characterize the local sample point cloud data corresponding to the sample grid; Then use the pixel information of the pixels to determine the label information of the sample grid that matches the pixels, which can reduce the difficulty of labeling and improve the labeling speed.
  • determining the annotation label information of the sample grid that matches each pixel based on the pixel information of each pixel in the top view of the sample includes:
  • the label information of adjacent pixels of the target pixel is adjusted to the label information of the target pixel.
  • the traffic markings of each category have a certain width
  • the width and adjusting the label information of adjacent pixels of the target pixel by expanding the width and adjusting the label information of adjacent pixels of the target pixel, the accuracy and accuracy of the label information of adjacent pixels can be improved. rationality.
  • an embodiment of the present disclosure also provides a device for identifying traffic markings, including:
  • a dividing module used to rasterize the road point cloud data to obtain local point cloud data contained in at least one target grid
  • a first determination module configured to determine the local point cloud data contained in each target grid. The probability that the target raster belongs to the traffic marking category;
  • the second determination module is used to determine the marking line recognition result of the road point cloud data based on the probability of the traffic marking category to which each of the target grids respectively belongs.
  • the traffic marking category includes a lane line category and a road boundary line category;
  • the first determination module based on the local points included in each target grid, Cloud data, when determining the probability that the target grid belongs to the traffic marking category, for the local point cloud data contained in each target grid, for each point cloud in the local point cloud data Perform feature extraction on point cloud information of points to generate road feature information of the target grid;
  • a first probability that the target grid belongs to the lane line category and a second probability that the target grid belongs to the road boundary line category are determined.
  • the second determination module determines the marking recognition result of the road point cloud data based on the probability of the traffic marking category to which each of the target grids respectively belongs. when, for each of the target grids, determine whether there is a target probability greater than a preset threshold in the first probability and the second probability of the target grid;
  • a marking line identification result of the road point cloud data is determined.
  • the second determination module is configured to determine the recognition result of the target grid based on the traffic marking category associated with the target probability.
  • the probability includes the first probability and the second probability, determine the maximum probability among the first probability and the second probability;
  • the traffic marking category associated with the maximum probability is used as the identification result of the target grid.
  • the road point cloud data is collected by a driving device, and the device further includes:
  • a control module configured to control the driving of the driving device based on the marking information of at least one traffic marking indicated by the marking line recognition result after determining the marking line recognition result of the road point cloud data, wherein,
  • the reticle information includes reticle position and/or reticle category.
  • the first determination module determines the probability that the target grid belongs to the traffic marking category based on the local point cloud data contained in each target grid. is used to use the trained target neural network to determine the probability that the target grid belongs to the traffic marking category based on the local point cloud data contained in each target grid.
  • the device further includes:
  • the training module is used to train and obtain the target neural network according to the following steps:
  • the neural network to be trained is iteratively trained until the training cutoff condition is met, and the result is obtained Describe the target neural network.
  • the predicted probability includes a first predicted probability that the sample raster belongs to the lane line category, and a second predicted probability that the sample raster belongs to the road boundary line category;
  • the training module performs iterative training on the neural network to be trained based on the predicted probability of the traffic marking category to which each of the sample grids respectively belongs and the annotation label information of each of the sample grids, Until the training cutoff condition is met and the target neural network is obtained, the method is used to, when the annotation label information of the sample grid indicates that the sample grid does not belong to the background, the method based on the sample grid
  • the first prediction probability, the second prediction probability, the label information of the sample grid, and the label information corresponding to the background label determine the first loss;
  • the label information of the sample grid indicates that the sample grid belongs to the background, based on the first predicted probability, the second predicted probability and the label information corresponding to the background label, it is determined second loss;
  • the neural network to be trained is iteratively trained until a training cutoff condition is met, and the target neural network is obtained.
  • the first loss includes a first sub-loss and a second sub-loss
  • the training module determines the first prediction probability, the second prediction probability based on the sample grid, the label information of the sample grid, and the label information corresponding to the background label. When there is a loss, it is used to determine, from the first predicted probability and the second predicted probability of the sample grid, a target that matches the traffic marking category indicated by the label information of the sample grid. predicted probability;
  • the second sub-loss is determined based on other prediction probabilities in the first prediction probability and the second prediction probability except the target prediction probability, and annotation label information corresponding to the background label.
  • the training module when determining the annotation label information of each sample grid, is used to based on the local sample point cloud data contained in each of the sample grids, Generate a top view of the sample; wherein each sample grid corresponds to a pixel in the top view of the sample;
  • the annotation label information of the sample grid matching each pixel is determined.
  • the training module determines the annotation label information of the sample grid that matches each pixel based on the pixel information of each pixel in the top view of the sample. , used to determine the annotation label information corresponding to each pixel in the sample top view based on the pixel information of each pixel in the sample top view;
  • the label information of adjacent pixels of the target pixel is adjusted to the label information of the target pixel.
  • an optional implementation manner of the present disclosure also provides a computer device, a processor, and a memory.
  • the memory stores machine-readable instructions executable by the processor, and the processor is configured to execute the instructions stored in the memory.
  • Machine-readable instructions when the machine-readable instructions are executed by the processor, when the machine-readable instructions are executed by the processor, the above-mentioned first aspect, or any possible implementation of the first aspect, is executed. steps in the way.
  • an optional implementation manner of the present disclosure also provides a computer-readable storage medium.
  • the computer-readable storage medium stores a computer program. When the computer program is run, it executes the above-mentioned first aspect, or any of the first aspects. steps in a possible implementation.
  • Figure 1 shows a flow chart of a method for identifying traffic markings provided by an embodiment of the present disclosure
  • Figure 2 shows a flow chart of a method for training a neural network to be trained provided by an embodiment of the present disclosure
  • Figure 3 shows a schematic diagram of a traffic marking recognition device provided by an embodiment of the present disclosure
  • FIG. 4 shows a schematic structural diagram of a computer device provided by an embodiment of the present disclosure.
  • the present disclosure provides a traffic marking recognition solution. Since the road point cloud data captured by lidar has morphological stability, using the acquired road point cloud data to identify the traffic marking category can improve the identification. Continuity of traffic markings. By rasterizing the road point cloud data, and then processing the local point cloud data corresponding to each target raster obtained after rasterization, the probability that each target raster belongs to the traffic marking category can be determined.
  • each target grid belonging to the traffic marking category can be accurately determined from multiple target grids, and then each target grid belonging to the traffic marking category can be used to accurately determine the road
  • Each point cloud point in the point cloud data belongs to the traffic marking category, and then the traffic markings of each category are accurately obtained, that is, the marking recognition result.
  • the execution subject of the method for identifying traffic markings provided by an embodiment of the disclosure generally has certain computing capabilities.
  • Terminal equipment or other processing equipment where the terminal equipment can be user equipment (User Equipment, UE), mobile equipment, user terminals, terminals, personal digital processing equipment (Personal Digital Assistant, PDA), handheld devices, computer equipment, etc.; in In some possible implementations, the traffic marking recognition method can be implemented by the processor calling computer readable instructions stored in the memory.
  • a flow chart of a method for identifying traffic markings may include the following steps:
  • the road point cloud data can be obtained using a lidar installed on the driving device.
  • the road point cloud data can be a point cloud vector set, and the point cloud vector set includes point cloud information of multiple point cloud points.
  • the point cloud information may include the coordinates of the point cloud points in the three-dimensional world coordinate system, the color information of the point cloud points, reflection intensity information, distance information, etc.
  • the laser radar installed on the driving device can be used to collect road point cloud data of the driving road.
  • S102 Rasterize the road point cloud data to obtain local point cloud data contained in at least one target raster.
  • the road point cloud data is rasterized to obtain multiple rasters, wherein the multiple rasters obtained may include empty rasters and non-empty target rasters.
  • Each target raster can include local point cloud data.
  • the local point cloud data is part of the point cloud data in the road point cloud data, including at least one point cloud point in the road point cloud data, and point cloud information of each point cloud point in the at least one point cloud point.
  • the local point cloud data contained in each target raster can form the above-mentioned road point cloud data.
  • the horizontal axis direction in the world coordinate system can be the direction of the road
  • the vertical axis direction (i.e., the y-axis direction) in the world coordinate system can be perpendicular to the direction of the road.
  • the vertical axis direction (i.e., the z-axis direction) can be the direction perpendicular to the road and pointing to the sky (or the ground).
  • the road point cloud data can be rasterized and divided according to the preset grid size.
  • the preset grid size may be L meters * M meters * N meters, where L is the length in the x-axis direction, M is the length in the y-axis direction, and N is the length in the z-axis direction. length.
  • the default grid size can be 0.16m*0.16m*15m.
  • the values corresponding to L, M, and N can be determined according to the parameters of the actual lidar, and are not specifically limited in the embodiments of the present disclosure. For example, it can be determined based on the maximum x value, maximum y value, and maximum z value among the coordinates of the point cloud points in the road point cloud data acquired by the lidar.
  • the number of divided target grids and the location of each point cloud point can be determined according to the preset grid size and the coordinates of each point cloud point in the road point cloud data in the world coordinate system.
  • the target raster uses each point cloud point located in the same target raster as the local point cloud data contained in the target raster.
  • At least one target grid can be obtained, and the local point cloud data contained in each target grid in the at least one target grid can be obtained.
  • S103 Based on the local point cloud data contained in each target grid, determine the probability that the target grid belongs to the traffic marking category.
  • the traffic markings may specifically be any markings on the road.
  • traffic markings can be lane lines, road boundary lines, zebra crossings, turn indicators, etc.
  • the probability that the target raster belongs to the traffic marking can be determined based on the point cloud information of each point cloud point in the local point cloud data contained in the target raster. . For example, based on the point cloud information of each point cloud point of the target grid, it can be determined whether each point cloud point belongs to the traffic marking line. According to the number of point cloud points belonging to the traffic marking line in the target grid, accounting for The ratio of the total number of point cloud points in the target grid determines the probability that the target grid belongs to the traffic marking category.
  • the first number of point cloud points belonging to the traffic markings in the target grid and the first number of point cloud points belonging to the background in the target grid can be determined.
  • the second quantity determines the probability that the target grid belongs to the traffic marking category based on the ratio of the first quantity and the second quantity.
  • the traffic marking category may include a lane line category and a road boundary line category, that is, the traffic markings may include traffic markings of the lane line category and traffic markings of the road boundary line category.
  • the traffic markings of the road boundary line category include road boundary lines located at the edges of both sides of the road; the traffic markings of the lane line category include various traffic markings on the road other than the traffic markings of the non-road boundary line category.
  • S103-1 For the local point cloud data contained in each target grid, perform feature extraction on the point cloud information of each point cloud point in the local point cloud data to generate road feature information of the target grid.
  • the road feature information is high-dimensional feature information, for example, 64-dimensional, 128-dimensional, etc.
  • Road feature information is feature information that can characterize whether the target raster is related to the traffic marking category in the road.
  • feature extraction can be performed on the point cloud information of each point cloud point in the local point cloud data contained in the target raster, and the point cloud information of each point cloud point can be extracted.
  • target feature information related to the traffic marking category For example, for each point cloud point, the first feature information related to the lane line category can be extracted from the point cloud information of the point cloud point, and, from the point cloud information of the point cloud point, the first feature information related to the lane line category can be extracted. Second feature information related to the road boundary line category. Then, the first feature information and the second feature information can be used as the target feature information of the point cloud point.
  • the target feature information corresponding to each point cloud point can be feature fused to obtain the road feature information corresponding to the target grid.
  • S103-2 Based on the road feature information, determine the first probability that the target raster belongs to the lane line category and the second probability that the target grid belongs to the road boundary line category.
  • the probability that the target grid belongs to the traffic marking category may include a first probability and a second probability.
  • the first probability is used to represent the probability that the point cloud points in the target grid belong to the lane line category
  • the second probability is used to represent the point cloud points in the target grid that belong to the road boundary line category. The probability.
  • the road feature information corresponding to the target grid can be convolved, and based on the results of the convolution process, the first probability that the target grid belongs to the lane line category and the second probability that the target grid belongs to the road boundary line category are determined.
  • the above S103 can be executed using the trained target neural network.
  • the local point cloud data contained in each target raster can be input into the trained target neural network in a serial manner, using The target neural network processes the local point cloud data contained in each target raster separately, and outputs the first probability that the target raster belongs to the lane line category and the second probability that the target raster belongs to the road boundary line category.
  • inputting to the trained target neural network for processing in series can reduce the processing pressure of the target neural network and make the target neural network more lightweight.
  • the feature extractor in the target neural network to extract the local point cloud data.
  • Feature extraction is performed on the point cloud information of each point cloud point in the data to obtain high-dimensional road feature information of the target raster, and then feature processing is performed on the road feature information to output the first probability sum that the target raster belongs to the lane line category.
  • the second probability of belonging to the road boundary line category may include two parts, one part is used to extract road feature information, and the other part is used to perform feature processing on the road feature information, and output the first probability and the second probability.
  • the part used to extract road feature information can include a layer of fully connected layer, a layer of batch normalization layer, a layer of linear rectification function (ReLU) function layer and a layer of max pooling layer.
  • the local point cloud data contained in the target raster can be input to the fully connected layer, and the point cloud information of each point cloud point in the local point cloud data can be fully connected to obtain the first point cloud point of each point cloud point.
  • intermediate feature information then input the first intermediate feature information of each point cloud point into the batch normalization layer, convert the range of feature values corresponding to each first intermediate feature information, and obtain the first intermediate feature information of each point cloud point.
  • Second intermediate feature information then input the second intermediate feature information of each point cloud point into the ReLU function layer, use the ReLU function to numerically transform the eigenvalues corresponding to each second intermediate feature information, and set the eigenvalues less than 0 to is 0, retain the feature values greater than 0, thereby obtaining the third intermediate feature information of each point cloud point; finally, the third intermediate feature information of each point cloud point can be input to the maximum pooling layer, using the maximum pooling The layer fuses the third intermediate feature information of each point cloud point to obtain high-dimensional road feature information of the target grid.
  • the part of the feature extractor used to output the first probability and the second probability may also include multiple network layers, where the multiple network layers may be a 2-dimensional (2D) convolution layer, a batch normalization layer, and a ReLU.
  • Function layer, upsampling layer and sigmoid activation function layer, multiple network layers form a fully convolutional network.
  • the sigmoid function is used for the output of hidden layer neurons, and the value range is (0, 1). It can map a value to the interval of (0, 1) and can be used for binary classification.
  • the road feature information can be input into the 2D convolution layer, and the road feature information can be 2D convolved to obtain the fourth intermediate feature information; then the fourth intermediate feature information can be input into the batch
  • the unified layer converts the range of eigenvalues corresponding to the fourth intermediate feature information to obtain the fifth intermediate feature information; inputs the fifth intermediate feature information to the ReLU function layer, and uses the ReLU function to calculate the features corresponding to the fifth intermediate feature information.
  • the value is numerically converted to obtain the sixth intermediate feature information; the sixth intermediate feature information is input to the upsampling layer, and the upsampling layer is used to upsample the sixth intermediate feature information to obtain the seventh intermediate feature information; finally, the seventh intermediate feature information is The intermediate feature information is input to the sigmoid activation function layer, and the sigmoid activation function is used to perform feature processing on the seventh intermediate feature information, and output the first probability map that the target raster belongs to the lane line category and the second probability map that belongs to the road boundary line category. Among them, the probability intervals corresponding to the first probability map and the second probability map are both (0, 1). According to the first probability map, the first probability that the target grid belongs to the lane line category is obtained, and based on the second probability map, the second probability that the target grid belongs to the road boundary line category is obtained.
  • the trained target neural network has reliable prediction accuracy.
  • the probability that the target grid belongs to each category of traffic markings can be accurately determined.
  • the traffic marking recognition method provided by the embodiment of the present disclosure can also be directly executed using the trained target neural network. That is, the trained target neural network can be used to execute the above S101 to S103 and In the following S104, the trained target neural network is used to directly output the reticle recognition result.
  • S104 Determine the marking recognition result of the road point cloud data based on the probability of the traffic marking category to which each target grid belongs.
  • the marking recognition result is used to indicate the marking information of at least one traffic marking corresponding to the road point cloud data.
  • the mark recognition result can indicate the marking information of each lane corresponding to the road point cloud data and/or The marking information of each road boundary line corresponding to the road point cloud data.
  • the marking information may include the marking position and/or the marking category of the traffic marking.
  • the probability corresponding to the target grid will also only include one, and then it can be obtained from multiple Among the target grids, the marking grids whose probability is greater than the preset probability are screened out.
  • each reticle grid with connectivity is determined, and then based on the coordinates of the point cloud points in each reticle grid with connectivity, A traffic marking is determined, and the marking position of the traffic marking can be determined based on the coordinates of the point cloud points in each connected marking grid.
  • each traffic marking line corresponding to the road point cloud data can be determined, as well as the marking line information, that is, the marking line recognition result is obtained.
  • the probability corresponding to the target grid will also include the probability corresponding to each traffic marking category.
  • Probability For example, when the traffic marking category includes the lane line category and the road boundary line category, the probability corresponding to the target grid may include a first probability that the target grid belongs to the lane line category and a second probability that the target grid belongs to the road boundary line category.
  • the category probability corresponding to the traffic marking category can be filtered out from the probability corresponding to each target grid. For each target grid, it can be determined whether the category probability corresponding to the target grid is greater than the preset probability. If so, the traffic marking category indicated by the category probability is used as the traffic marking category to which the target grid belongs. If not, it is determined that the target grid does not belong to the traffic marking category. Based on this, the traffic marking category to which each target grid belongs can be determined.
  • each target grid with connectivity can be determined based on the position information of each target line grid in the world coordinate system, and then each target grid with connectivity can be determined based on the location information of each target line grid in the world coordinate system.
  • the coordinates of the point cloud points in the target grid determine a traffic marking. Based on the coordinates of the point cloud points in each target grid corresponding to the traffic marking, the marking position of the traffic marking is determined; at the same time, the traffic marking can be associated with the class probability of each target grid corresponding to the traffic marking Line category, as the marking category of the traffic marking. In this way, the traffic markings of each category corresponding to the road point cloud data and the marking information of each traffic marking can be obtained.
  • the road point cloud data captured by lidar has morphological stability
  • using the acquired road point cloud data to identify traffic marking categories can improve the continuity of the identified traffic markings.
  • the probability that each target raster belongs to the traffic marking category can be determined.
  • each target grid belonging to the traffic marking category can be accurately determined from multiple target grids, and then each target grid belonging to the traffic marking category can be used to accurately determine the road
  • Each point cloud point in the point cloud data belongs to the traffic marking category, and then the traffic markings of each category are accurately obtained, that is, the marking recognition result.
  • S104 when the traffic marking category includes the lane line category and the road boundary line category, and the probability of the traffic marking category to which the target grid belongs includes the first probability and the second probability, S104 may follow these steps to implement:
  • S104-1 For each target grid, determine whether there is a target probability greater than a preset threshold in the first probability and the second probability of the target grid.
  • the preset threshold is a preset minimum probability value
  • the lane line category and the road boundary line category may correspond to the same preset threshold.
  • the preset threshold can be set based on experience and is not specifically limited here.
  • the first probability is greater than the preset threshold, it can be determined that the target grid is more likely to belong to the traffic marking category associated with the first probability; similarly, when the second probability is greater than the preset threshold, It can be determined that the target grid is more likely to belong to the traffic marking category of the second probability association.
  • the target probability is the probability between the first probability and the second probability that is greater than the preset threshold.
  • the first probability and the second probability of the target grid can be compared with a preset threshold respectively, thereby determining whether there is a target probability in the first probability and the second probability. If yes, the following S104-1 may be performed. If not, that is, neither the first probability nor the second probability of the target raster is greater than the preset threshold, it can be determined that the target raster belongs to the background raster, that is, it can be determined that the local point cloud data contained in the target raster Each point cloud point of is a point cloud point of the background category.
  • the preset threshold may also include two, specifically a first preset threshold corresponding to the lane line category and a second preset threshold corresponding to the road boundary line category.
  • the first probability can be compared with the first preset threshold, and in the first If the probability is greater than the first preset threshold, it is determined that the first probability belongs to the target probability; otherwise, it is determined that the first probability does not belong to the target probability.
  • the second probability can be compared with the second preset threshold. If the second probability is greater than the second preset threshold, it is determined that the second probability belongs to the target probability; otherwise, it can be determined that the second probability does not belong to the target probability. .
  • the target grid belongs to the background grid, that is, it can be determined that the target grid contains Each point cloud point in the local point cloud data is a point cloud point of the background category.
  • S104-2 When there is a target probability, determine the recognition result corresponding to the target grid according to the traffic marking category associated with the target probability.
  • the traffic marking category associated with the target probability is the lane line category
  • the traffic marking category associated with the target probability is the road boundary. Line category.
  • the recognition result is used to indicate the traffic marking category to which the target raster belongs.
  • the recognition result corresponding to the target grid can be determined based on the traffic marking category associated with the target probability. For example, when the traffic marking category associated with the target probability is the lane line category, the corresponding recognition result of the target raster is: the target raster belongs to the traffic marking category, that is, the local point cloud data contained in the target raster Each point cloud point in is a point cloud point of the traffic marking category. For another example, when the traffic marking category associated with the target probability is the road boundary line category, the corresponding recognition result of the target raster is: the target raster belongs to the road boundary line category, that is, the local points contained in the target raster Each point cloud point in the cloud data is a point cloud point of the road boundary line category.
  • the target probability includes both the first probability and the second probability
  • the maximum probability can be screened out from the first probability and the second probability included in the target probability.
  • the traffic marking category associated with the maximum probability is used as the recognition result of the target grid. That is, the traffic marking category with the highest probability association is used as the traffic marking category to which the target raster belongs. In this way, the uniqueness of the obtained target probability can be guaranteed and the rationality of the obtained recognition results can be improved.
  • S104-3 Based on the recognition results of each target grid, determine the marking line recognition results of the road point cloud data.
  • each target grid belonging to the same traffic marking category can be determined based on the recognition results corresponding to each target grid. Then, each target grid with connectivity can be determined based on the coordinates of the point cloud points in the local point cloud data contained in each target grid belonging to the same traffic marking category. According to each target grid with connectivity, The coordinates of the point cloud points in the grid are used to determine a traffic marking with this traffic marking category. Moreover, the marking position of each traffic marking line can be determined based on the coordinates of the point cloud points in the target grid corresponding to each traffic marking line.
  • the traffic marking category to which the traffic marking belongs and the marking position of the traffic marking can be used as the marking information of the traffic marking.
  • the marking recognition results of the road point cloud data can be determined based on the marking information of each traffic marking.
  • the road point cloud data may be collected by the driving device.
  • the road point cloud data can be collected using a laser radar installed on the driving device.
  • the driving device can also be controlled to drive based on the marking line information of at least one traffic marking line indicated by the marking line recognition result, wherein the marking line information includes the marking line position and/or the marking line. Line category.
  • the driving device may include an autonomous vehicle, a manually driven vehicle, a robot, or any other device that can travel on the road.
  • the marking position is used to represent the position of the traffic marking in the world coordinate system.
  • the marking category is the traffic marking category to which the traffic marking belongs, for example, the lane line category and the road boundary line category.
  • the marking information may include but is not limited to the marking position and/or the marking category. For example, it may also include the distance between the traffic marking and the traveling device, the angle between the direction of the traffic marking and the traveling direction of the traveling device, etc.
  • each traffic marking indicated by the marking recognition result can be determined, as well as the marking position and marking type of each traffic marking.
  • the safe driving area is determined, and the driving device is controlled to drive in the safe driving area.
  • an alarm may be prompted. For example, voice alarm prompts, buzzer alarm prompts, etc.
  • the embodiment of the present disclosure also provides a method for training the to-be-trained neural network, as shown in Figure 2, which is an implementation of the present disclosure.
  • the example provides a flow chart of a method for training a neural network to be trained, which may include the following steps:
  • sample point cloud data can be road point cloud data collected using any lidar.
  • Sample point cloud data can be a sample point cloud vector set, and the sample point cloud vector set includes point cloud information of multiple sample point cloud points.
  • the point cloud information may include the coordinates of the sample point cloud points in the three-dimensional world coordinate system, color information, reflection intensity information, distance information, etc. of the sample point cloud points.
  • S202 Rasterize the sample point cloud data to obtain local sample point cloud data contained in at least one sample raster; and determine the annotation label information of each sample raster.
  • the annotation label information may specifically be the label value corresponding to the sample raster.
  • the annotation label information can be the first label value corresponding to the lane line category; when the sample raster belongs to the road boundary line category, the annotation label information can be the first label value corresponding to the road boundary line category.
  • the number of divided sample grids and the size of each sample point cloud can be determined according to the preset grid size and the coordinates of each sample point cloud point in the sample point cloud data in the world coordinate system.
  • the sample grid where the point is located takes each sample point cloud point located in the same sample grid as the local sample point cloud data contained in the sample grid.
  • the label information corresponding to the sample grid can be determined in advance.
  • annotation label information of each sample grid can be determined according to the following steps:
  • Step 1 Generate a sample top view based on the local sample point cloud data contained in each sample grid; where each sample grid corresponds to a pixel in the sample top view.
  • each sample grid can be projected separately to obtain a top view of the sample.
  • the number of pixels in the sample top view is the number of sample grids, and one sample grid corresponds to one pixel in the sample top view.
  • the pixel information of the pixel point can be determined based on the point cloud information of each sample point cloud point in the local sample point cloud data contained in the sample grid.
  • pixel point 1 corresponds to sample grid 1
  • the pixel information of pixel point 1 can be determined based on the point cloud information of each sample point cloud point in the local sample point cloud data contained in sample grid 1.
  • Step 2 Based on the pixel information of each pixel in the top view of the sample, determine the annotation label information of the sample grid that matches each pixel.
  • the sample grid matching the pixel point is the sample grid corresponding to the pixel point.
  • the semantic label of the pixel can be determined based on the pixel information of the pixel.
  • semantic labels can include lane line labels, road boundary line labels and background labels.
  • the lane line label represents the lane line category
  • the road boundary line label represents the road boundary line category
  • the background label represents the background category.
  • Different semantic labels correspond to different label values. Specifically, the lane line label corresponds to the first label value, the road boundary line label corresponds to the second label value, and the background label corresponds to the third label value.
  • the annotation label information of the sample grid that matches each pixel can be determined. For example, for any pixel, the label value corresponding to the semantic label of the pixel can be used as the annotation label information of the sample grid corresponding to the pixel.
  • step 2 can be implemented according to the following steps:
  • the semantic label of the pixel can be determined based on the pixel information of the pixel in a manual labeling manner, and then the label value corresponding to the semantic label of the pixel can be used as the pixel.
  • annotation label information The annotation label information of a pixel is the annotation label information corresponding to the pixel.
  • the traffic label category may include a lane line category and a road boundary line category.
  • the preset extension width can be determined based on the number of pixels that need to be used and are located on the left and right sides of the target pixel. For example, for any target pixel, if you need one pixel adjacent to the left and right sides of the target pixel (that is, the pixel adjacent to the left of the target pixel, and the adjacent pixel to the right of the target pixel) pixels) to adjust the label information, the default expansion width can be 3 pixels wide.
  • each target pixel point belonging to the traffic marking category can be determined based on the annotation label information of each pixel point. Then for each target pixel, the preset extension width can be used to determine two adjacent pixels located on the left and right sides of the target pixel and adjacent to the target pixel. Afterwards, the label information of two adjacent pixels can be adjusted to the label information of the target pixel. Here, if the label information of the adjacent pixels of the target pixel is the same as the label information of the target pixel, If the label information is the same, the label information of adjacent pixels does not need to be adjusted.
  • each adjacent pixel point in the top view of the sample with adjusted label information can be used as a sample grid that matches each adjacent pixel point with adjusted label information.
  • grid label information At the same time, in the top view of the sample, except for the adjacent pixels whose label information has been adjusted, the unadjusted label information of other pixels can be used as the label information of the sample grid that matches other pixels.
  • the other pixels may include unadjusted pixels labeled with label information.
  • each lane line and each road boundary line in the sample top view can be determined based on the semantic label of each pixel.
  • the corresponding width can be 1 pixel wide.
  • the preset expansion width can be used to expand the width of each lane line and each road boundary line. That is, for each pixel in each lane line, the adjacent pixels on the left and right sides of the pixel can be used as added pixels in the lane line, thereby expanding the width of each lane line and obtaining the expansion.
  • the annotation label information of the sample grid corresponding to each pixel in the expanded lane line can be set as the first label value corresponding to the lane line category.
  • the adjacent pixels on the left and right sides of the pixel can be used as additional pixels in the road boundary line, thereby expanding the width of each road boundary line and obtaining the expansion.
  • road boundary line behind the annotation label information of the sample grid corresponding to each pixel in the expanded road boundary line can be set as the second label value corresponding to the road boundary line category.
  • the annotation label information of the corresponding sample grid is set to the third label value corresponding to the background category.
  • S203 Input the local sample point cloud data contained in each sample grid to the neural network to be trained, and generate a predicted probability that the sample grid belongs to the traffic marking category.
  • the neural network to be trained is the target neural network to be trained.
  • the predicted probability is the probability that the sample raster output by the neural network to be trained belongs to the traffic marking category.
  • the local sample point cloud data contained in each sample raster can be input to the neural network to be trained, and the local sample point cloud data can be processed using the neural network to be trained.
  • the output sample raster belongs to the traffic marking. The predicted probability of the category.
  • S204 Based on the predicted probability of the traffic marking category to which each sample grid belongs and the annotation label information of each sample grid, iteratively train the neural network to be trained until the training cutoff condition is met, and the target neural network is obtained.
  • the training cutoff condition may include that the number of rounds of iterative training reaches a preset number of rounds, and/or the prediction accuracy of the trained neural network reaches a preset accuracy.
  • the prediction loss of the neural network to be trained can be determined based on the predicted probability that the sample raster belongs to the traffic marking category and the label information of the sample raster, and the prediction loss is used to iteratively train the neural network to be trained until the training requirements are met. Cutoff conditions to obtain the target neural network.
  • the predicted probability may include a first predicted probability that the sample raster belongs to the lane line category, and a second predicted probability that the sample raster belongs to the road boundary line category.
  • the first prediction based on the sample grid can be The first loss is determined based on the probability, the second predicted probability, the label information of the sample grid, and the label information corresponding to the background label.
  • the first loss may include a first sub-loss and a second sub-loss.
  • the step of determining the first loss may be implemented according to the following sub-steps:
  • Sub-step 1 Determine the target predicted probability that matches the traffic marking category indicated by the label information of the sample grid from the first predicted probability and the second predicted probability of the sample grid.
  • the target prediction probability is the first prediction probability corresponding to the sample grid, and the label value corresponding to the annotation label information is the first label value; in the annotation label
  • the target prediction probability is corresponding to the sample raster
  • the second prediction probability of , the label value corresponding to the label information is the second label value.
  • the traffic marking category to which the sample grid actually belongs can be determined based on the label information of the sample grid. Then, from the first predicted probability and the second predicted probability of the sample grid, a target predicted probability that matches the traffic marking category to which the sample grid actually belongs is determined.
  • Sub-step 2 Determine the first sub-loss based on the target prediction probability and the annotation label information of the sample grid.
  • the binary cross-entropy function can be used to determine the first sub-loss based on the target prediction probability and the label value corresponding to the annotation label information.
  • Sub-step 3 Determine the second sub-loss based on the other predicted probabilities in the first predicted probability and the second predicted probability except the target predicted probability, and the annotation label information corresponding to the background label.
  • the background label is the background category
  • the annotation label information corresponding to the background label is the third label value.
  • the binary cross-entropy function can be used to determine the second sub-loss based on the third label value corresponding to other predicted probabilities and the annotation label information of the background label.
  • the target prediction probability is the first prediction probability
  • the other prediction probabilities are the second prediction probability.
  • the binary cross-entropy function can be used to determine the first sub-loss based on the first prediction probability and the first label value.
  • the binary cross-entropy function can be used to determine the second sub-loss based on the second prediction probability and the third label value.
  • the target prediction probability is the second prediction probability
  • the other prediction probabilities are the first prediction probability.
  • the binary cross-entropy function can be used to determine the first sub-loss based on the second prediction probability and the second label value.
  • the binary cross-entropy function can be used to determine the second sub-loss based on the first prediction probability and the third label value.
  • first sub-loss and second sub-loss can be regarded as the first loss.
  • the first prediction probability and the second prediction probability can be used.
  • the annotation label information corresponding to the background label determines the second loss.
  • the second loss may include a third sub-loss and a fourth sub-loss.
  • the binary cross entropy function can be used to determine the third sub-loss based on the first prediction probability and the third label value corresponding to the sample grid; and the binary cross entropy function can be used to determine the third sub-loss based on the second prediction probability corresponding to the sample grid. Predict the probability and the third label value to determine the fourth sub-loss.
  • the third sub-loss and the fourth sub-loss as the second loss.
  • the neural network to be trained may be iteratively trained until the training cutoff condition is met, and the target neural network is obtained.
  • the writing order of each step does not mean a strict execution order and does not constitute any limitation on the implementation process.
  • the specific execution order of each step should be based on its function and possible The internal logic is determined.
  • the embodiments of the present disclosure also provide a traffic marking identification device corresponding to the traffic marking identification method. Since the problem-solving principle of the device in the embodiments of the present disclosure is the same as that of the traffic markings in the embodiments of the present disclosure, The identification methods are similar, so the implementation of the device can be referred to the implementation of the method, and repeated details will not be repeated.
  • a schematic diagram of a traffic marking recognition device provided by an embodiment of the present disclosure includes:
  • Acquisition module 301 is used to obtain road point cloud data
  • the dividing module 302 is used to rasterize the road point cloud data to obtain local point cloud data contained in at least one target grid;
  • the first determination module 303 is configured to determine the probability that the target grid belongs to the traffic marking category based on the local point cloud data contained in each target grid;
  • the second determination module 304 is configured to determine the marking line recognition result of the road point cloud data based on the probability of the traffic marking category to which each of the target grids respectively belongs.
  • the traffic marking category includes a lane line category and a road boundary line category;
  • the first determination module 303 based on the local information contained in each target grid, Point cloud data, when determining the probability that the target grid belongs to the traffic marking category, is used for the local point cloud data contained in each target grid, for each point in the local point cloud data Feature extraction is performed on the point cloud information of the cloud points to generate road feature information of the target grid;
  • the second determination module 304 determines the marking identification of the road point cloud data based on the probability of the traffic marking category to which each of the target grids respectively belongs. As a result, for each of the target grids, determine whether there is a target probability greater than a preset threshold in the first probability and the second probability of the target grid;
  • a marking line identification result of the road point cloud data is determined.
  • the second determination module 304 is configured to determine the recognition result of the target grid according to the traffic marking category associated with the target probability.
  • the target probability includes the first probability and the second probability, determine the maximum probability among the first probability and the second probability;
  • the traffic marking category associated with the maximum probability is used as the identification result of the target grid.
  • the road point cloud data is collected by a driving device, and the device further includes:
  • the control module 305 is configured to control the driving of the driving device based on the marking information of at least one traffic marking indicated by the marking line recognition result after the determination of the marking line recognition result of the road point cloud data, wherein , the reticle information includes reticle position and/or reticle category.
  • the first determination module 303 determines that the target grid belongs to the traffic marking category based on the local point cloud data contained in each target grid.
  • the trained target neural network is used to determine the probability that the target grid belongs to the traffic marking category based on the local point cloud data contained in each target grid.
  • the device further includes:
  • the training module 306 is used to train and obtain the target neural network according to the following steps:
  • the neural network to be trained is iteratively trained until the training cutoff condition is met, and the result is obtained Describe the target neural network.
  • the predicted probability includes a first predicted probability that the sample raster belongs to the lane line category, and a second predicted probability that the sample raster belongs to the road boundary line category;
  • the training module 306 performs iterative training on the neural network to be trained based on the predicted probability of the traffic marking category to which each of the sample grids respectively belongs and the annotation label information of each of the sample grids. , until the training cutoff condition is met and the target neural network is obtained, for when the annotation label information of the sample grid indicates that the sample grid does not belong to the background, based on all the parameters of the sample grid.
  • the first prediction probability, the second prediction probability, the label information of the sample grid, and the label information corresponding to the background label are used to determine the first loss;
  • the label information of the sample grid indicates that the sample grid belongs to the background, based on the first predicted probability, the second predicted probability and the label information corresponding to the background label, it is determined second loss;
  • the neural network to be trained is iteratively trained until a training cutoff condition is met, and the target neural network is obtained.
  • the first loss includes a first sub-loss and a second sub-loss
  • the training module 306 determines based on the first predicted probability, the second predicted probability of the sample grid, the label information of the sample grid, and the label information corresponding to the background label.
  • the first loss is used to determine, from the first predicted probability and the second predicted probability of the sample grid, the traffic marking category that matches the traffic marking category indicated by the label information of the sample grid. target prediction probability;
  • the second sub-loss is determined based on other prediction probabilities in the first prediction probability and the second prediction probability except the target prediction probability, and annotation label information corresponding to the background label.
  • the training module 306, when determining the annotation label information of each sample grid is configured to base on the local sample point cloud data contained in each of the sample grids. , generate a top view of the sample; wherein each sample grid corresponds to a pixel in the top view of the sample;
  • the annotation label information of the sample grid matching each pixel is determined.
  • the training module 306 determines the annotation label information of the sample grid that matches each pixel based on the pixel information of each pixel in the top view of the sample. When, it is used to determine the annotation label information corresponding to each pixel in the sample top view based on the pixel information of each pixel in the sample top view;
  • the label information of adjacent pixels of the target pixel is adjusted to the label information of the target pixel.
  • a schematic structural diagram of a computer device provided by an embodiment of the present application includes:
  • Processor 41 memory 42 and bus 43.
  • the memory 42 stores machine-readable instructions executable by the processor 41
  • the processor 41 is used to execute the machine-readable instructions stored in the memory 42 .
  • the processor 41 executes The following steps: S101: Obtain road point cloud data; S102: Rasterize the road point cloud data to obtain local point cloud data contained in at least one target raster; S103: Based on the data contained in each target raster Based on the local point cloud data, determine the probability that the target grid belongs to the traffic marking category and S104: Based on the probability of the traffic marking category to which each target grid belongs, determine the marking recognition result of the road point cloud data.
  • the above-mentioned memory 42 includes a memory 421 and an external memory 422; the memory 421 here is also called an internal memory, and is used to temporarily store the operation data in the processor 41, as well as the data exchanged with external memory 422 such as a hard disk.
  • the processor 41 communicates with the processor 42 through the memory 421.
  • the external memory 422 performs data exchange.
  • the processor 41 and the memory 42 communicate through the bus 43, so that the processor 41 executes the execution instructions mentioned in the above method embodiment.
  • Embodiments of the present disclosure also provide a computer-readable storage medium.
  • a computer program is stored on the computer-readable storage medium.
  • the computer program executes the traffic marking identification method described in the above method embodiment. step.
  • the storage medium may be a volatile or non-volatile computer-readable storage medium.
  • the computer program product of the traffic marking recognition method provided by the embodiment of the present disclosure includes a computer-readable storage medium storing program code.
  • the instructions included in the program code can be used to execute the traffic marking method described in the above method embodiment.
  • the computer program product can be implemented specifically through hardware, software or a combination thereof.
  • the computer program product is embodied as a computer storage medium.
  • the computer program product is embodied as a software product, such as a Software Development Kit (SDK), etc. wait.
  • SDK Software Development Kit
  • the disclosed devices and methods can be implemented in other ways.
  • the device embodiments described above are only illustrative.
  • the division of the units is only a logical function division.
  • multiple units or components may be combined.
  • some features can be ignored, or not implemented.
  • the coupling or direct coupling or communication connection between each other shown or discussed may be through some communication interfaces, and the indirect coupling or communication connection of the devices or units may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separated.
  • the components shown as units may or may not be physical units, that is, they may be located in one place, or they may be distributed to multiple on the network unit. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of this embodiment.
  • each functional unit in various embodiments of the present disclosure may be integrated into one processing unit, or each unit may exist physically alone, or two or more units may be integrated into one unit.
  • the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a non-volatile computer-readable storage medium that is executable by a processor.
  • the technical solution of the present disclosure is essentially or the part that contributes to the existing technology or the part of the technical solution can be embodied in the form of a software product.
  • the computer software product is stored in a storage medium, including Several instructions are used to cause a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present disclosure.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM), random access memory (Random Access Memory, RAM), magnetic disk or optical disk and other media that can store program code. .
  • the products applying the technical solution of this application will clearly inform the personal information processing rules and obtain the individual's independent consent before processing personal information.
  • the product applying the technical solution in this application must obtain the individual's separate consent before processing sensitive personal information, and meet the requirement of "express consent" at the same time. For example, setting up clear and conspicuous signs on personal information collection devices such as cameras to inform them that they have entered the scope of personal information collection, and that personal information will be collected.
  • personal information processing rules may include personal information processing rules.
  • Information processors purposes of personal information processing, processing methods, types of personal information processed, etc.

Landscapes

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

Abstract

本公开提供了一种交通标线的识别方法、装置、计算机设备和存储介质，其中，该方法包括：获取道路点云数据；对道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；基于每个目标栅格所包含的局部点云数据，确定目标栅格属于交通标线类别的概率；基于各个目标栅格分别所属的交通标线类别的概率，确定道路点云数据的标线识别结果。

Description

交通标线的识别方法、装置、计算机设备和存储介质
相关申请的交叉引用
本专利申请要求于2022年7月1日提交的、申请号为202210773244.8的中国专利申请的优先权，该申请以引用的方式并入文本中。
技术领域
本公开涉及计算机视觉技术领域，具体而言，涉及一种交通标线的识别方法、装置、计算机设备和存储介质。
背景技术
随着计算机视觉技术的快速发展，计算机视觉技术在越来越多的领域中得到应用。在自动驾驶技术领域中，通常通过设置在车辆上的摄像装置拍摄道路图像，再基于道路图像，确定出道路中的车道线、道路边界线等交通标线距离车辆的距离。
但是，基于道路图像确定出的车道线、道路边界线，通常存在线段不连续、不准确的问题。
发明内容
本公开实施例至少提供一种交通标线的识别方法、装置、计算机设备和存储介质。
第一方面，本公开实施例提供了一种交通标线的识别方法，包括：
获取道路点云数据；
对所述道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；
基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率；
基于各个所述目标栅格分别所属的交通标线类别的概率，确定所述道路点云数据的标线识别结果。
该实施方式，由于利用激光雷达拍摄的道路点云数据具有形态稳定性，利用获取的道路点云数据进行交通标线类别的识别，可以提高识别出的交通标线的连续性。通过对道路点云数据进行栅格化划分，再对栅格化划分后得到的各个目标栅格对应的局部点云数据(即各个目标栅格所包含的局部点云数据)进行处理，可以确定出每个目标栅格属于交通标线类别的概率，基于各个目标栅格分别所属的交通标线类别的概率，可以从多个目标栅格中，准确确定出属于交通标线类别的各个目标栅格，再利用属于交通标线类别的各个目标栅格，能够准确确定出道路点云数据中属于交通标线类别的各个点云点，进而准确得到各个类别的交通标线，即标线识别结果。
在一种可能的实施方式中，所述交通标线类别包括车道线类别和道路边界线类别；所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率，包括：
针对每个所述目标栅格所包含的所述局部点云数据，对所述局部点云数据中每个点云点的点云信息进行特征提取，生成所述目标栅格的道路特征信息；
基于所述道路特征信息，确定所述目标栅格属于所述车道线类别的第一概率和属于所述道路边界线类别的第二概率。
该实施方式，通过对局部点云数据中每个点云点的点云信息进行特征提取，可以充分提取出各个点云点中与交通标线类别相关的特征信息，继而得到目标栅格的准确道路特征信息。利用道路特征信息，确定目标栅格属于车道线类别的第一概率和属于道路边界线类别的第二概率，可以实现对目标栅格对应所属交通标线类别的准确划分，进而得到目标栅格的准确交通标线类别。
在一种可能的实施方式中，所述基于各个所述目标栅格分别所属的交通标线类别的 概率，确定所述道路点云数据的标线识别结果，包括：
针对每个所述目标栅格，确定所述目标栅格的所述第一概率和所述第二概率中是否存在大于预设阈值的目标概率；
在存在所述目标概率的情况下，根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果；
基于各个所述目标栅格的所述识别结果，确定所述道路点云数据的标线识别结果。
该实施方式，利用预设阈值，可以筛选出概率值较大的目标概率，目标概率关联的交通标线类别，为目标栅格对应所属的交通标线类别的可能性更大，从而，基于目标概率关联的交通标线类别，确定出的识别结果也更准确。如此，利用各个目标概率分别确定出各个目标栅格的识别结果，再基于各个识别结果确定标线识别结果，可以提高确定出的标线识别结果的准确性。
在一种可能的实施方式中，所述根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果，包括：
在所述目标概率包括所述第一概率和所述第二概率的情况下，确定所述第一概率和所述第二概率中的最大概率；
将所述最大概率关联的所述交通标线类别，作为所述目标栅格的识别结果。
该实施方式，在目标概率包括第一概率和第二概率的情况下，将最大概率关联的交通标线类别，作为目标栅格的识别结果，可以保障得到的目标概率的唯一性，提高得到的识别结果的合理性。
在一种可能的实施方式中，所述道路点云数据为行驶装置采集的，在所述确定所述道路点云数据的标线识别结果之后，还包括：
基于所述标线识别结果指示的至少一条交通标线的标线信息，控制所述行驶装置行驶，其中，所述标线信息包括标线位置和/或标线类别。
该实施方式，通过确定的标线位置和/或标线类别，控制行驶装置行驶，可以保障行驶装置在合理的区域行驶，提高行驶装置的行驶安全性。
在一种可能的实施方式中，所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率，包括：
利用训练后的目标神经网络，基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率。
该实施方式，训练后的目标神经网络具有可靠的预测精度，利用训练后的目标神经网络，可以准确确定出目标栅格属于各个类别的交通标线的概率。
在一种可能的实施方式中，根据下述步骤训练得到所述目标神经网络：
获取样本点云数据；
对所述样本点云数据进行栅格化划分，得到至少一个样本栅格所包含的局部样本点云数据；并确定每个样本栅格的标注标签信息；
将每个所述样本栅格所包含的所述局部样本点云数据输入至待训练神经网络，生成所述样本栅格属于所述交通标线类别的预测概率；
基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
该实施方式，通过样本栅格属于交通标线类别的预测概率和每个样本栅格的标注标签信息，对待训练神经网络进行迭代训练，可以提高输出的预测概率和标注标签信息之间的一致性，从而得到具有可靠预测精度的目标神经网络。
在一种可能的实施方式中，所述预测概率包括所述样本栅格属于车道线类别的第一预测概率、和属于道路边界线类别的第二预测概率；
所述基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络，包括：
在所述样本栅格的所述标注标签信息指示所述样本栅格不属于背景的情况下，基于 所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失；
在所述样本栅格的所述标注标签信息指示所述样本栅格属于背景的情况下，基于所述第一预测概率、所述第二预测概率和所述背景标签对应的标注标签信息，确定第二损失；
基于所述第一损失和所述第二损失中的至少一种，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
该实施方式，通过确定各个损失，再利用确定出的各个损失对待训练神经网络进行迭代训练，可以提高网络输出的预测概率和标注标签信息之间的一致性，从而得到具有可靠预测精度的目标神经网络。
在一种可能的实施方式中，所述第一损失包括第一子损失和第二子损失；
所述基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失，包括：
从所述样本栅格的所述第一预测概率和所述第二预测概率中，确定出与所述样本栅格的标注标签信息指示的交通标线类别相匹配的目标预测概率；
基于所述目标预测概率和所述样本栅格的标注标签信息，确定第一子损失；
基于所述第一预测概率和所述第二预测概率中除所述目标预测概率之外的其他预测概率，和背景标签对应的标注标签信息，确定第二子损失。
该实施方式，第一子损失能够表征网络输出的预测概率和标注标签信息之间的差异，第二子损失能够表征网络输出的预测概率和背景标签对应的标注标签信息之间的差异，利用第一子损失和第二子损失对网络进行训练，能够提高网络输出的预测概率和标注标签信息之间的一致性。
在一种可能的实施方式中，所述确定每个样本栅格的标注标签信息，包括：
基于各个所述样本栅格分别所包含的所述局部样本点云数据，生成样本俯视图；其中，每个所述样本栅格对应所述样本俯视图中的一个像素点；
基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
该实施方式，基于各个样本栅格分别所包含的局部样本点云数据，生成样本俯视图，可以实现利用样本俯视图中的像素点的像素信息，对样本栅格对应的局部样本点云数据进行表征；再利用像素点的像素信息，确定与像素点匹配的样本栅格的标注标签信息，能够降低标注难度，提高标注速度。
在一种可能的实施方式中，所述基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息，包括：
基于所述样本俯视图中各个像素点的像素信息，确定所述样本俯视图中每个像素点对应的标注标签信息；
针对所述标注标签信息指示为交通标线类别的目标像素点，基于预设的扩展宽度，将所述目标像素点的相邻像素点的标注标签信息调整为所述目标像素点的标注标签信息；
基于所述样本俯视图中所述相邻像素点的调整后的标注标签信息、和除所述相邻像素点之外的其他像素点的未调整的所述标注标签信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
该实施方式，由于各个类别的交通标线具有一定的宽度，通过扩展宽度，对目标像素点的相邻像素点的标注标签信息进行调整，可以提高相邻像素点的标注标签信息的准确性和合理性。
第二方面，本公开实施例还提供一种交通标线的识别装置，包括：
获取模块，用于获取道路点云数据；
划分模块，用于对所述道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；
第一确定模块，用于基于每个所述目标栅格所包含的所述局部点云数据，确定所述 目标栅格属于交通标线类别的概率；
第二确定模块，用于基于各个所述目标栅格分别所属的交通标线类别的概率，确定所述道路点云数据的标线识别结果。
在一种可能的实施方式中，所述交通标线类别包括车道线类别和道路边界线类别；所述第一确定模块，在所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率时，用于针对每个所述目标栅格所包含的所述局部点云数据，对所述局部点云数据中每个点云点的点云信息进行特征提取，生成所述目标栅格的道路特征信息；
基于所述道路特征信息，确定所述目标栅格属于所述车道线类别的第一概率和属于所述道路边界线类别的第二概率。
在一种可能的实施方式中，所述第二确定模块，在所述基于各个所述目标栅格分别所属的交通标线类别的所述概率，确定所述道路点云数据的标线识别结果时，用于针对每个所述目标栅格，确定所述目标栅格的所述第一概率和所述第二概率中是否存在大于预设阈值的目标概率；
在存在所述目标概率的情况下，根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果；
基于各个所述目标栅格的所述识别结果，确定所述道路点云数据的标线识别结果。
在一种可能的实施方式中，所述第二确定模块，在所述根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果时，用于在所述目标概率包括所述第一概率和所述第二概率的情况下，确定所述第一概率和所述第二概率中的最大概率；
将所述最大概率关联的所述交通标线类别，作为所述目标栅格的识别结果。
在一种可能的实施方式中，所述道路点云数据为行驶装置采集的，所述装置还包括：
控制模块，用于在所述确定所述道路点云数据的标线识别结果之后，基于所述标线识别结果指示的至少一条交通标线的标线信息，控制所述行驶装置行驶，其中，所述标线信息包括标线位置和/或标线类别。
在一种可能的实施方式中，所述第一确定模块，在所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率时，用于利用训练后的目标神经网络，基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率。
在一种可能的实施方式中，所述装置还包括：
训练模块，用于根据下述步骤训练得到所述目标神经网络：
获取样本点云数据；
对所述样本点云数据进行栅格化划分，得到至少一个样本栅格所包含的局部样本点云数据；并确定每个样本栅格的标注标签信息；
将每个所述样本栅格所包含的所述局部样本点云数据输入至待训练神经网络，生成所述样本栅格属于所述交通标线类别的预测概率；
基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
在一种可能的实施方式中，所述预测概率包括所述样本栅格属于车道线类别的第一预测概率、和属于道路边界线类别的第二预测概率；
所述训练模块，在所述基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络时，用于在所述样本栅格的所述标注标签信息指示所述样本栅格不属于背景的情况下，基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失；
在所述样本栅格的所述标注标签信息指示所述样本栅格属于背景的情况下，基于所述第一预测概率、所述第二预测概率和所述背景标签对应的标注标签信息，确定第二损失；
基于所述第一损失和所述第二损失中的至少一种，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
在一种可能的实施方式中，所述第一损失包括第一子损失和第二子损失；
所述训练模块，在所述基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失时，用于从所述样本栅格的所述第一预测概率和所述第二预测概率中，确定出与所述样本栅格的标注标签信息指示的交通标线类别相匹配的目标预测概率；
基于所述目标预测概率和所述样本栅格的标注标签信息，确定第一子损失；
基于所述第一预测概率和所述第二预测概率中除所述目标预测概率之外的其他预测概率，和背景标签对应的标注标签信息，确定第二子损失。
在一种可能的实施方式中，所述训练模块，在所述确定每个样本栅格的标注标签信息时，用于基于各个所述样本栅格分别所包含的所述局部样本点云数据，生成样本俯视图；其中，每个所述样本栅格对应所述样本俯视图中的一个像素点；
基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
在一种可能的实施方式中，所述训练模块，在所述基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息时，用于基于所述样本俯视图中各个像素点的像素信息，确定所述样本俯视图中每个像素点对应的标注标签信息；
针对所述标注标签信息指示为交通标线类别的目标像素点，基于预设的扩展宽度，将所述目标像素点的相邻像素点的标注标签信息调整为所述目标像素点的标注标签信息；
基于所述样本俯视图中所述相邻像素点的调整后的标注标签信息、和除所述相邻像素点之外的其他像素点的未调整的所述标注标签信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
第三方面，本公开可选实现方式还提供一种计算机设备，处理器、存储器，所述存储器存储有所述处理器可执行的机器可读指令，所述处理器用于执行所述存储器中存储的机器可读指令，所述机器可读指令被所述处理器执行时，所述机器可读指令被所述处理器执行时执行上述第一方面，或第一方面中任一种可能的实施方式中的步骤。
第四方面，本公开可选实现方式还提供一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被运行时执行上述第一方面，或第一方面中任一种可能的实施方式中的步骤。
关于上述交通标线的识别装置、计算机设备、及计算机可读存储介质的效果描述参见上述交通标线的识别方法的说明，这里不再赘述。
为使本公开的上述目的、特征和优点能更明显易懂，下文特举较佳实施例，并配合所附附图，作详细说明如下。
附图说明
为了更清楚地说明本公开实施例的技术方案，下面将对实施例中所需要使用的附图作简单地介绍，此处的附图被并入说明书中并构成本说明书中的一部分，这些附图示出了符合本公开的实施例，并与说明书一起用于说明本公开的技术方案。应当理解，以下附图仅示出了本公开的某些实施例，因此不应被看作是对范围的限定，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相关的附图。
图1示出了本公开实施例所提供的一种交通标线的识别方法的流程图；
图2示出了本公开实施例所提供的一种对待训练神经网络进行训练的方法的流程图；
图3示出了本公开实施例所提供的一种交通标线的识别装置的示意图；
图4示出了本公开实施例所提供的一种计算机设备的结构示意图。
具体实施方式
为使本公开实施例的目的、技术方案和优点更加清楚，下面将结合本公开实施例中附图，对本公开实施例中的技术方案进行清楚、完整地描述，显然，所描述的实施例仅仅是本公开一部分实施例，而不是全部的实施例。通常在此处描述和示出的本公开实施例的组件可以以各种不同的配置来布置和设计。因此，以下对本公开的实施例的详细描述并非旨在限制要求保护的本公开的范围，而是仅仅表示本公开的选定实施例。基于本公开的实施例，本领域技术人员在没有做出创造性劳动的前提下所获得的所有其他实施例，都属于本公开保护的范围。
另外，本公开实施例中的说明书和权利要求书及上述附图中的术语“第一”、“第二”等是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。
在本文中提及的“多个或者若干个”是指两个或两个以上。“和/或”，描述关联对象的关联关系，表示可以存在三种关系，例如，A和/或B，可以表示：单独存在A，同时存在A和B，单独存在B这三种情况。字符“/”一般表示前后关联对象是一种“或”的关系。
经研究发现，在利用摄像装置拍摄的道路图像进行交通标线识别时，通过需要将道路图像中属于交通标线的像素点，在图像坐标系中的坐标转换为世界坐标系中的坐标，再基于像素点对应在世界坐标系中的坐标进行交通标线的确定。这种方式下，不仅增加了交通标线确定的复杂性，还将引入坐标转换的误差，导致确定出的交通标线不准确、不连续。
基于上述研究，本公开提供了一种交通标线的识别方案，由于利用激光雷达拍摄的道路点云数据具有形态稳定性，利用获取的道路点云数据进行交通标线类别的识别，可以提高识别出的交通标线的连续性。通过对道路点云数据进行栅格化划分，再对栅格化划分后得到的各个目标栅格对应的局部点云数据进行处理，可以确定出每个目标栅格属于交通标线类别的概率，基于各个目标栅格对应的概率，可以从多个目标栅格中，准确确定出属于交通标线类别的各个目标栅格，再利用属于交通标线类别的各个目标栅格，能够准确确定出道路点云数据中属于交通标线类别的各个点云点，进而准确得到各个类别的交通标线，即标线识别结果。
针对以上方案所存在的缺陷，均是发明人在经过实践并仔细研究后得出的结果，因此，上述问题的发现过程以及下文中本公开针对上述问题所提出的解决方案，都应该是发明人在本公开过程中对本公开做出的贡献。
应注意到：相似的标号和字母在下面的附图中表示类似项，因此，一旦某一项在一个附图中被定义，则在随后的附图中不需要对其进行进一步定义和解释。
为便于对本实施例进行理解，首先对本公开实施例所公开的一种交通标线的识别方法进行详细介绍，本公开实施例所提供的交通标线的识别方法的执行主体一般为具有一定计算能力的终端设备或其他处理设备，其中终端设备可以为用户设备(User Equipment，UE)、移动设备、用户终端、终端、个人数字处理设备(Personal Digital Assistant，PDA)、手持设备、计算机设备等；在一些可能的实现方式中，该交通标线的识别方法可以通过处理器调用存储器中存储的计算机可读指令的方式来实现。
下面以执行主体为计算机设备为例对本公开实施例提供的交通标线的识别方法加以说明。
如图1所示，为本公开实施例提供的一种交通标线的识别方法的流程图，可以包括以下步骤：
S101：获取道路点云数据。
这里，道路点云数据可以利用安装在行驶装置上的激光雷达获取，道路点云数据可以为点云向量集合，点云向量集合中包括多个点云点的点云信息。其中，点云信息可以包括点云点在三维的世界坐标系中的坐标、点云点的颜色信息、反射强度信息、距离信息等。
示例性的，在行驶装置行驶的过程中，可以利用安装在行驶装置上的激光雷达，采集行驶道路的道路点云数据。
S102：对道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据。
这里，对道路点云数据进行栅格化划分，可以得到多个栅格，其中，得到的多个栅格中可以包括空的栅格和非空的目标栅格。每个目标栅格可以包括局部点云数据。
局部点云数据为道路点云数据中的部分点云数据，包括道路点云数据中至少一个点云点，以及至少一个点云点中的每个点云点的点云信息。各个目标栅格分别所包含的局部点云数据，可以组成上述道路点云数据。
示例性的，世界坐标系中的横轴方向(即x轴方向)可以为道路所在的方向，世界坐标系中的纵轴方向(即y轴方向)可以垂直道路所在方向，世界坐标系中的竖轴方向(即z轴方向)可以为垂直道路所在方向、且指向天空(或地面)的方向。
在获取到道路点云数据之后，可以按照预设栅格尺寸，对道路点云数据进行栅格化划分。示例性的，预设栅格尺寸可以为L米*M米*N米，其中，L为在x轴方向上的长度，M为在y轴方向上的长度，N为在z轴方向上的长度。例如，预设栅格尺寸可以为0.16米*0.16米*15米。
在具体应用时，L、M、N分别对应的数值，可以根据实际使用的激光雷达的参数确定，本公开实施例不进行具体限定。例如，可以根据激光雷达获取的道路点云数据中的点云点的坐标中的、最大x值、最大y值和最大z值确定。
具体实施时，可以按照预设栅格尺寸，以及道路点云数据中每个点云点在世界坐标系中的坐标，确定划分出的目标栅格的数量，以及每个点云点所位于的目标栅格，将位于同一目标栅格内的各个点云点，作为该目标栅格所包含的局部点云数据。
基于对道路点云数据进行栅格化划分，可以得到至少一个目标栅格，以及至少一个目标栅格中每个目标栅格所包含的局部点云数据。
S103：基于每个目标栅格所包含的局部点云数据，确定目标栅格属于交通标线类别的概率。
这里，交通标线具体可以为道路中的任一标线。例如，交通标线可以为车道线、道路边界线、斑马线、转向指示线等。
具体实施时，针对划分后得到的每个目标栅格，可以基于该目标栅格所包含的局部点云数据中的各个点云点的点云信息，确定该目标栅格属于交通标线的概率。示例性的，根据目标栅格的每个点云点的点云信息，可以确定出每个点云点是否属于交通标线，根据目标栅格中属于交通标线的点云点的数量，占目标栅格中总点云点数量的比值，确定目标栅格属于交通标线类别的概率。
或者，可以根据目标栅格的每个点云点的点云信息，确定出目标栅格中属于交通标线的点云点的第一数量，和目标栅格中属于背景的点云点的第二数量，基于第一数量和第二数量的比值，确定目标栅格属于交通标线类别的概率。
在一种实施例中，交通标线类别可以包括车道线类别和道路边界线类别，也即，交通标线可以包括车道线类别的交通标线和道路边界线类别的交通标线。其中，道路边界线类别的交通标线包括位于道路两侧边缘处的道路边界线；车道线类别的交通标线为道路中，非道路边界线类别的交通标线以外的各种交通标线。
针对上述S103，可以按照以下步骤实施：
S103-1：针对每个目标栅格所包含的局部点云数据，对局部点云数据中每个点云点的点云信息进行特征提取，生成目标栅格的道路特征信息。
这里，道路特征信息为高维的特征信息，例如，64维、128维等。道路特征信息为能够表征目标栅格与道路中的交通标线类别是否相关的特征信息。
具体实施时，针对每个目标栅格，可以对该目标栅格所包含的局部点云数据中每个点云点的点云信息进行特征提取，提取出每个点云点的点云信息中、与交通标线类别相关的目标特征信息。例如，针对每个点云点，可以从该点云点的点云信息中，提取出与车道线类别相关的第一特征信息，以及，从该点云点的点云信息中，提取出与道路边界线类别相关的第二特征信息。然后可以将第一特征信息和第二特征信息，作为该点云点的目标特征信息。
之后，可以再对各个点云点分别对应的目标特征信息进行特征融合，得到目标栅格对应的道路特征信息。
S103-2：基于道路特征信息，确定目标栅格属于车道线类别的第一概率和属于道路边界线类别的第二概率。
这里，在交通标线类别包括车道线类别和道路边界线类别的情况下，目标栅格属于交通标线类别的概率可以包括第一概率和第二概率。其中，第一概率用于表征目标栅格中的点云点属于车道线类别的点云点的概率；第二概率用于表征目标栅格中的点云点属于道路边界线类别的点云点的概率。
具体实施时，可以对目标栅格对应的道路特征信息进行卷积处理，基于卷积处理的结果，确定出目标栅格属于车道线类别的第一概率和属于道路边界线类别的第二概率。
在一种实施例中，上述S103可以利用训练后的目标神经网络执行。具体实施时，在获取到每个目标栅格所包含的局部点云数据之后，可以将各个目标栅格分别所包含的局部点云数据，以串联的方式输入至训练后的目标神经网络，利用目标神经网络分别对每个目标栅格所包含的局部点云数据进行处理，输出目标栅格属于车道线类别的第一概率和属于道路边界线类别的第二概率。这样，通过串联的方式输入至训练后的目标神经网络进行处理，可以降低目标神经网络的处理压力，使得目标神经网络可以更轻量。
示例性的，针对每个目标栅格，在将该目标栅格所包含的局部点云数据输入至训练后的目标神经网络后，可以先利用目标神经网络中的特征提取器，对局部点云数据中的每个点云点的点云信息进行特征提取，得到目标栅格的高维的道路特征信息，再对道路特征信息进行特征处理，输出目标栅格属于车道线类别的第一概率和属于道路边界线类别的第二概率。具体的，特征提取器可以包括两部分，一部分用于提取道路特征信息，一部分用于对道路特征信息进行特征处理，输出第一概率和第二概率。
其中，用于提取道路特征信息的部分，可以包括一层全连接层、一层批归一化层、一层线性整流函数(Linear rectification function，简称ReLU)函数层和一层最大池化层。具体的，可以将目标栅格所包含的局部点云数据输入至全连接层，对局部点云数据中每个点云点的点云信息进行全连接处理，得到每个点云点的第一中间特征信息；再将每个点云点的第一中间特征信息输入至批归一化层，对每个第一中间特征信息对应的特征值的范围进行转换，得到每个点云点的第二中间特征信息；再将每个点云点的第二中间特征信息输入至ReLU函数层，利用ReLU函数对每个第二中间特征信息对应的特征值进行数值转化，将小于0的特征值置为0，将大于0的特征值保留，从而得到每个点云点的第三中间特征信息；最后，可以将每个点云点的第三中间特征信息输入至最大池化层，利用最大池化层将各个点云点的第三中间特征信息进行特征融合，得到目标栅格的高维的道路特征信息。
特征提取器中用于输出第一概率和第二概率的部分，也可以包括多个网络层，其中，多个网络层可以分别为2维(2D)卷积层、批归一化层、ReLU函数层、上采样层和sigmoid激活函数层，多个网络层构成全卷积网络。其中，sigmoid函数用于隐层神经元输出，取值范围为(0，1)，它可以将一个数值映射到(0，1)的区间，可以用来做二分类。
在得到目标栅格的道路特征信息后，可以将道路特征信息输入中2D卷积层，对道路特征信息进行2D卷积，得到第四中间特征信息；再将第四中间特征信息输入至批归一化层，对第四中间特征信息对应的特征值的范围进行转换，得到第五中间特征信息；将第五中间特征信息输入至ReLU函数层，利用ReLU函数对第五中间特征信息对应的特征值进行数值转化，得到第六中间特征信息；将第六中间特征信息输入至上采样层，利用上采样层对第六中间特征信息进行上采样处理，得到第七中间特征信息；最后，将第七中间特征信息输入至sigmoid激活函数层，利用sigmoid激活函数对第七中间特征信息进行特征处理，输出目标栅格属于车道线类别的第一概率图和属于道路边界线类别的第二概率图。其中，第一概率图和第二概率图对应的概率区间均为(0，1)。根据第一概率图，也即得到目标栅格属于车道线类别的第一概率，根据第二概率图，也即得到目标栅格属于道路边界线类别的第二概率。
这样，训练后的目标神经网络具有可靠的预测精度，利用训练后的目标神经网络，可以准确确定出目标栅格属于各个类别的交通标线的概率。
在一种可能的方式中，本公开实施例提供的交通标线的识别方法，也可以直接利用训练后的目标神经网络执行，也即，可以训练后的目标神经网络，执行上述S101～S103以及下述S104，利用训练后的目标神经网络，直接输出标线识别结果。
S104：基于各个目标栅格分别所属的交通标线类别的概率，确定道路点云数据的标线识别结果。
这里，标线识别结果用于指示道路点云数据对应的至少一条交通标线的标线信息，具体的，标识识别结果可以指示道路点云数据对应的各条车道线的标线信息和/或道路点云数据对应的各条道路边界线的标线信息。标线信息可以包括交通标线的标线位置和/或标线类别。
示例性的，在得到各个目标栅格分别所属的交通标线类别的概率之后，在交通标线类别仅包括一个的情况下，目标栅格对应的概率也将只包括一个，然后可以从多个目标栅格中，筛选出概率大于预设概率的标线栅格。基于筛选出的每个标线栅格在世界坐标系下的位置信息，确定具有联通性的各个标线栅格，然后可以根据具有联通性的各个标线栅格中的点云点的坐标，确定出一条交通标线，并可以基于具有联通性的各个标线栅格中的点云点的坐标，确定该条交通标线的标线位置。如此，基于各个目标栅格分别对应的概率，可以确定道路点云数据对应的各条交通标线，以及标线信息，也即得到了标线识别结果。
示例性的，在得到各个目标栅格分别所属的交通标线类别的概率之后，在交通标线类别包括多个的情况下，目标栅格对应的概率也将包括各个交通标线类别分别对应的概率。例如，在交通标线类别包括车道线类别和道路边界线类别的情况下，目标栅格对应的概率可以包括目标栅格属于车道线类别的第一概率和属于道路边界线类别的第二概率。在交通标线类别包括多个的情况下，针对每个交通标线类别，可以从每个目标栅格对应的概率中，筛选出该交通标线类别对应的类别概率。针对每个目标栅格，可以确定该目标栅格对应的类别概率是否大于预设概率，若是，则将该类别概率指示的交通标线类别作为该目标栅格所属的交通标线类别。若否，则确定该目标栅格不属于该交通标线类别。基于此，可以确定出每个目标栅格所属的交通标线类别。
之后，针对属于同一交通标线类别的各个目标栅格，可以基于每个目标线栅格在世界坐标系下的位置信息，确定具有联通性的各个目标栅格，然后可以根据具有联通性的各个目标栅格中的点云点的坐标，确定出一条交通标线。基于该交通标线对应的各个目标栅格中的点云点的坐标，确定该条交通标线的标线位置；同时可以将该交通标线对应的各个目标栅格的类别概率关联的交通标线类别，作为该交通标线的标线类别。如此，可以得到道路点云数据对应的各个类别的交通标线，以及各个交通标线的标线信息。
这样，由于利用激光雷达拍摄的道路点云数据具有形态稳定性，利用获取的道路点云数据进行交通标线类别的识别，可以提高识别出的交通标线的连续性。通过对道路点云数据进行栅格化划分，再对栅格化划分后得到的各个目标栅格对应的局部点云数据进行处理，可以确定出每个目标栅格属于交通标线类别的概率，基于各个目标栅格对应的概率，可以从多个目标栅格中，准确确定出属于交通标线类别的各个目标栅格，再利用属于交通标线类别的各个目标栅格，能够准确确定出道路点云数据中属于交通标线类别的各个点云点，进而准确得到各个类别的交通标线，即标线识别结果。
在一种实施例中，针对S104，在交通标线类别包括车道线类别和道路边界线类别，目标栅格所属的交通标线类别的概率包括第一概率和第二概率的情况下，S104可以按照以下步骤实施：
S104-1：针对每个目标栅格，确定目标栅格的第一概率和第二概率中是否存在大于预设阈值的目标概率。
这里，预设阈值为预先设定的最小概率值，车道线类别和道路边界线类别可以对应于同一个预设阈值。预设阈值可以根据经验进行设置，此处不进行具体限定。在第一概率大于该预设阈值的情况下，可以确定目标栅格属于第一概率关联的交通标线类别的可能性较大；同样的，在第二概率大于该预设阈值的情况下，可以确定目标栅格属于第二概率关联的交通标线类别的可能性较大。
目标概率为第一概率和第二概率中大于预设阈值的概率。
具体实施时，针对每个目标栅格，可以将该目标栅格的第一概率和第二概率分别与预设阈值进行比较，从而确定出第一概率和第二概率中是否存在目标概率。如果是，则可以执行下述S104-1。如果否，也即目标栅格的第一概率和第二概率均不大于预设阈值，则可以确定目标栅格属于背景栅格，也即，可以确定目标栅格所包含的局部点云数据中的各个点云点均为背景类别的点云点。
在一种实施例中，预设阈值也可以包括两个，具体为车道线类别对应的第一预设阈值、和道路边界线类别对应的第二预设阈值。
针对上述S104-1中“确定目标栅格的第一概率和第二概率中是否存在大于预设阈值的目标概率”的步骤，可以将第一概率和第一预设阈值进行比较，在第一概率大于第一预设阈值的情况下，确定第一概率属于目标概率；反之，确定第一概率不属于目标概率。同时，可以将第二概率和第二预设阈值进行比较，在第二概率大于第二预设阈值的情况下，确定第二概率属于目标概率；反之，则可以确定第二概率不属于目标概率。
这样，通过为不同的道路标线类别设置不同的预设阈值，再利用各个道路标线类别对应的预设阈值，进行目标概率的筛选，可以提高确定出的目标概率的合理性和准确性。
另外，在第一概率不大于第一预设阈值、且第二概率不大于第二预设阈值的情况下，则可以确定目标栅格属于背景栅格，也即，可以确定目标栅格所包含的局部点云数据中的各个点云点均为背景类别的点云点。
S104-2：在存在目标概率的情况下，根据目标概率关联的交通标线类别，确定目标栅格对应的识别结果。
这里，在目标概率为第一概率的情况下，目标概率关联的交通标线类别即为车道线类别；在目标概率为第二概率的情况下，目标概率关联的交通标线类别即为道路边界线类别。识别结果用于指示目标栅格所属的交通标线类别。
具体实施时，可以根据目标概率关联的交通标线类别，确定目标栅格对应的识别结果。例如，在目标概率关联的交通标线类别为车道线类别的情况下，则目标栅格对应的识别结果为：目标栅格属于交通标线类别，也即目标栅格所包含的局部点云数据中的各个点云点均为交通标线类别的点云点。又例如，在目标概率关联的交通标线类别为道路边界线类别的情况下，则目标栅格对应的识别结果为：目标栅格属于道路边界线类别，也即目标栅格所包含的局部点云数据中的各个点云点均为道路边界线类别的点云点。
在一种实施例中，在利用车道线类别和道路边界线类别对应的同一个预设阈值筛选目标概率，或者利用第一预设阈值和第二预设阈值筛选目标概率的过程中，均存在目标概率既包括第一概率又包括第二概率的情况，在这种情况下，针对上述S104-2，可以从目标概率包括的第一概率和第二概率中，筛选出最大概率。然后，将最大概率关联的交通标线类别，作为目标栅格的识别结果。也即，将最大概率关联的交通标线类别，作为目标栅格所属的交通标线类别。这样，可以保障得到的目标概率的唯一性，提高得到的识别结果的合理性。
S104-3：基于各个目标栅格的识别结果，确定道路点云数据的标线识别结果。
具体实施时，可以根据各个目标栅格分别对应的识别结果，确定属于同一交通标线类别的各个目标栅格。然后，可以根据属于同一交通标线类别的各个目标栅格，分别所包含的局部点云数据中的点云点的坐标，确定具有联通性的各个目标栅格，根据具有联通性的各个目标栅格中的点云点的坐标，确定具有该交通标线类别的一条交通标线。并且，可以基于各条交通标线分别对应的目标栅格中的点云点的坐标，确定各条交通标线的标线位置。
然后，针对每条交通标线，可以将该条交通标线所属的交通标线类别、该条交通标线的标线位置，作为该条交通标线的标线信息。最后，可以根据每条交通标线的标线信息，确定出道路点云数据的标线识别结果。
在一种实施例中，道路点云数据可以为行驶装置采集的。具体的，道路点云数据可以为利用安装在行驶装置上的激光雷达采集的。在得到道路点云数据的标线识别结果之后，还可以基于标线识别结果指示的至少一条交通标线的标线信息，控制行驶装置行驶，其中，标线信息包括标线位置和/或标线类别。
这里，行驶装置可以包括自动驾驶车辆、人工驾驶车辆、机器人等任一可以在道路上行驶的装置。标线位置用于表征交通标线在世界坐标系中的位置。标线类别即为交通标线所属的交通标线类别，例如，车道线类别和道路边界线类别。标线信息可以包括但不限于标线位置和/或标线类别，例如还可以包括交通标线距离行驶装置的距离、交通标线方向与行驶装置行驶方向之间的角度等。
示例性的，在得到标线识别结果之后，可以确定出标线识别结果指示的各条交通标线，以及各条交通标线的标线位置和标线类别。根据各条交通标线的标线位置和标线类别，确定安全行驶区域，并控制行驶装置在安全行驶区域中行驶。
又例如，在根据行驶装置的位置和道路边界线的标线位置，确定行驶装置与道路边界线之间的距离小于预设距离的情况下，可以进行报警提示。例如进行语音报警提示、蜂鸣报警提示等。
这样，通过确定的标线位置和/或标线类别，控制行驶装置行驶，可以保障行驶装置在合理的区域行驶，提高行驶装置的行驶安全性。
另外，由上述实施例可知，上述S103可以利用训练后的目标神经网络执行，因此，本公开实施例还提供了一种对待训练神经网络进行训练的方法，如图2所示，为本公开实施例提供的一种对待训练神经网络进行训练的方法的流程图，可以包括以下步骤：
S201：获取样本点云数据。
这里，样本点云数据可以为利用任一激光雷达采集的道路点云数据。样本点云数据 中可以为样本点云向量集合，样本点云向量集合中包括多个样本点云点的点云信息。其中，点云信息可以包括样本点云点在三维的世界坐标系中的坐标、样本点云点的颜色信息、反射强度信息、距离信息等。
S202：对样本点云数据进行栅格化划分，得到至少一个样本栅格所包含的局部样本点云数据；并确定每个样本栅格的标注标签信息。
这里，标注标签信息具体可以为样本栅格对应的标签值。在样本栅格属于车道线类别的情况下，标注标签信息可以为车道线类别对应的第一标签值；在样本栅格属于道路边界线类别的情况下，标注标签信息可以为道路边界线类别对应的第二标签值；在样本栅格属于背景类别的情况下，标注标签信息可以为背景类别对应的第三标签值。
具体实施时，可以按照预设栅格尺寸，以及样本点云数据中每个样本点云点的在世界坐标系中的坐标，确定出划分出的样本栅格的数量，以及每个样本点云点所位于的样本栅格，将位于同一样本栅格内的各个样本点云点，作为该样本栅格所包含的局部样本点云数据。同时，针对每个样本栅格，可以预先确定出该样本栅格对应的标注标签信息。
在一种实施例中，可以按照以下步骤确定每个样本栅格的标注标签信息：
步骤一、基于各个样本栅格分别所包含的局部样本点云数据，生成样本俯视图；其中，每个样本栅格对应样本俯视图中的一个像素点。
示例性的，在对样本点云数据进行栅格化划分，得到各个样本栅格之后，可以对各个样本栅格分别进行投影，得到样本俯视图。其中，样本俯视图中像素点的数量即为样本栅格的数量，一个样本栅格对应样本俯视图中的一个像素点。其中，像素点的像素信息可以根据样本栅格所包含的局部样本点云数据中的各个样本点云点的点云信息确定。例如，像素点1对应于样本栅格1，像素点1的像素信息可以根据样本栅格1所包含的局部样本点云数据中的各个样本点云点的点云信息确定。
步骤二、基于样本俯视图中各个像素点的像素信息，确定与每个像素点匹配的样本栅格的标注标签信息。
这里，与像素点匹配的样本栅格即为像素点对应的样本栅格。
具体实施时，在得到样本俯视图之后，针对样本俯视图中的每个像素点，可以基于该像素点的像素信息，确定该像素点的语义标签。其中，语义标签可以包括车道线标签、道路边界线标签和背景标签。
其中，车道线标签即表示车道线类别、道路边界线标签即表示道路边界线类别、背景标签即表示背景类别。不同的语义标签对应于不同的标签值，具体的，车道线标签对应于第一标签值，道路边界线标签对应于第二标签值，背景标签对应于第三标签值。
之后，可以基于各个像素点的语义标签对应的标签值，确定与各个像素点分别匹配的样本栅格的标注标签信息。例如，针对任一像素点，可以将该像素点的语义标签对应的标签值，作为该像素点对应的样本栅格的标注标签信息。
在一种实施例中，针对上述步骤二，可以按照以下步骤实施：
S1：基于样本俯视图中各个像素点的像素信息，确定样本俯视图中每个像素点对应的标注标签信息。
具体实施时，针对每个像素点，可以以人工标注的方式，根据该像素点的像素信息，确定该像素点的语义标签，然后可以将该像素点的语义标签对应的标签值作为该像素点的标注标签信息。像素点的标注标签信息即为像素点对应的标注标签信息。
S2：针对标注标签信息指示为交通标线类别的目标像素点，基于预设的扩展宽度，将目标像素点的相邻像素点的标注标签信息调整为目标像素点的标注标签信息。
这里，交通标签类别可以包括车道线类别和道路边界线类别。预设的扩展宽度可以根据需要使用的、分别位于目标像素点左右两侧的像素点的数量确定。例如，针对任一目标像素点，如果需要对该目标像素点左右两侧分别相邻的一个像素点(也即，目标像素点左侧相邻的像素点，和目标像素点右侧相邻的像素点)，进行标注标签信息的调整，则预设的扩展宽度可以为3像素宽。
具体实施时，在得到各个像素点的标注标签信息之后，可以根据每个像素点的标注标签信息，确定出属于交通标线类别的各个目标像素点。然后针对每个目标像素点，可以利用预设的扩展宽度，确定出位于该目标像素点左右侧、且与该目标像素点相邻的两个相邻像素点。之后，可以将两个相邻像素点的标注标签信息调整为该目标像素点的标注标签信息。这里，若目标像素点的相邻像素点的标注标签信息与目标像素点的标注标 签信息相同，则可以不对相邻像素点的标注标签信息进行调整。
S3：基于样本俯视图中相邻像素点的调整后的标注标签信息、和除相邻像素点之外的其他像素点的未调整的标注标签信息，确定与每个像素点匹配的样本栅格的标注标签信息。
具体实施时，可以将样本俯视图中调整了标注标签信息的各个相邻像素点，分别对应的调整后的标注标签信息，作为与调整了标注标签信息的各个相邻像素点，分别匹配的样本栅格的标注标签信息。同时，可以将样本俯视图中，除了调整了标注标签信息的各个相邻像素点之外的、其他像素点的未调整的标注标签信息，作为与其他像素点匹配的样本栅格的标注标签信息。其中，其他像素点可以包括未调整的标注标签信息的像素点。
在一种可能的实施方式，在得到样本俯视图之后，可以基于各个像素点的语义标签，确定出样本俯视图中的各条车道线和各条道路边界线。之后，针对任一条车道线或任一条道路边界线，其对应的宽度可以为1像素宽。之后，可以利用预设的扩展宽度，对每条车道线，以及每条道路边界线进行宽度扩展。也即，针对每条车道线中的各个像素点，可以将该像素点左右两侧的相邻像素点作为该车道线中的添加像素点，从而实现对每条车道线的宽度扩展，得到扩展后的车道线。之后，可以将扩展后的车道线中的每个像素点对应的样本栅格的标注标签信息，均设置为车道线类别对应的第一标签值。
针对每条道路边界线中的各个像素点，可以将该像素点左右两侧的相邻像素点作为该道路边界线中的添加像素点，从而实现对每条道路边界线的宽度扩展，得到扩展后的道路边界线。之后，可以将扩展后的道路边界线中的每个像素点对应的样本栅格的标注标签信息，均设置为道路边界线类别对应的第二标签值。
将不位于扩展后的车道线且不位于扩展后的道路边界线的各个像素点，分别对应的样本栅格的标注标签信息均设置为背景类别对应的第三标签值。
S203：将每个样本栅格所包含的局部样本点云数据输入至待训练神经网络，生成样本栅格属于交通标线类别的预测概率。
这里，待训练神经网络即为待训练的目标神经网络。预测概率即为待训练神经网络输出的、样本栅格属于交通标线类别的概率。
具体实施时，可以将依次将每个样本栅格所包含的局部样本点云数据输入至待训练神经网络，利用待训练神经网络对局部样本点云数据进行处理，输出样本栅格属于交通标线类别的预测概率。
S204：基于各个样本栅格分别所属的交通标线类别的预测概率、和每个样本栅格的标注标签信息，对待训练神经网络进行迭代训练，直至满足训练截止条件，得到目标神经网络。
这里，训练截止条件可以包括迭代训练的轮数达到预设轮数，和/或训练后的神经网络的预测精度达到预设精度。
具体实施时，可以根据样本栅格属于交通标线类别的预测概率，和样本栅格的标注标签信息，确定待训练神经网络的预测损失，利用预测损失对待训练神经网络进行迭代训练，直至满足训练截止条件，得到目标神经网络。
在一种实施例中，预测概率可以包括样本栅格属于车道线类别的第一预测概率、和属于道路边界线类别的第二预测概率。
具体实施时，针对上述S204，可以按照以下步骤实施：
在样本栅格的标注标签信息指示样本栅格不属于背景的情况下，也即在样本栅格的标注标签信息指示样本栅格不属于背景类别的情况下，可以基于样本栅格的第一预测概率、第二预测概率、样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失。
具体实施时，第一损失可以包括第一子损失和第二子损失，针对确定第一损失的步骤，可以按照如下子步骤实施：
子步骤一、从样本栅格的第一预测概率和第二预测概率中，确定出与样本栅格的标注标签信息指示的交通标线类别相匹配的目标预测概率。
其中，在标注标签信息指示的交通标线类别为车道线类别的情况下，目标预测概率为样本栅格对应的第一预测概率，标注标签信息对应的标签值为第一标签值；在标注标签信息指示的交通标线类别为道路边界线类别的情况下，目标预测概率为样本栅格对应 的第二预测概率，标注标签信息对应的标签值为第二标签值。
示例性的，针对每个样本栅格，可以根据该样本栅格的标注标签信息，确定该样本栅格实际所属的交通标线类别。之后，从样本栅格的第一预测概率和第二预测概率中，确定与该样本栅格实际所属的交通标线类别相匹的目标预测概率。
子步骤二、基于目标预测概率和样本栅格的标注标签信息，确定第一子损失。
具体实施时，可以利用二值交叉熵函数，基于目标预测概率和标注标签信息对应的标签值，确定第一子损失。
子步骤三、基于第一预测概率和第二预测概率中除目标预测概率之外的其他预测概率，和背景标签对应的标注标签信息，确定第二子损失。
这里，背景标签即背景类别，背景标签对应的标注标签信息即为第三标签值。具体的，可以利用二值交叉熵函数，基于其他预测概率和背景标签的标注标签信息对应的第三标签值，确定第二子损失。
示例性的，在样本栅格的标注标签信息指示的交通标线类别为车道线类别的情况下，目标预测概率即为第一预测概率，其他预测概率即为第二预测概率。之后，可以利用二值交叉熵函数，基于第一预测概率和第一标签值，确定第一子损失。同时，可以利用二值交叉熵函数，基于第二预测概率和第三标签值，确定第二子损失。
示例性的，在样本栅格的标注标签信息指示的交通标线类别为道路边界线类别的情况下，目标预测概率即为第二预测概率，其他预测概率即为第一预测概率。之后，可以利用二值交叉熵函数，基于第二预测概率和第二标签值，确定第一子损失。同时，可以利用二值交叉熵函数，基于第一预测概率和第三标签值，确定第二子损失。
最后，可以将得到的第一子损失和第二子损失，作为第一损失。
在样本栅格的标注标签信息指示样本栅格属于背景的情况下，也即在样本栅格的标注标签信息指示样本栅格属于背景类别的情况下，可以基于第一预测概率、第二预测概率和背景标签对应的标注标签信息，确定第二损失。
这里，第二损失可以包括第三子损失和第四子损失。具体实施时，可以利用二值交叉熵函数，基于样本栅格对应的第一预测概率和第三标签值，确定第三子损失；并利用二值交叉熵函数，基于样本栅格对应的第二预测概率和第三标签值，确定第四子损失。将第三子损失和第四子损失，作为第二损失。
进一步的，可以基于第一损失和第二损失中的至少一种，对待训练神经网络进行迭代训练，直至满足训练截止条件，得到目标神经网络。
本领域技术人员可以理解，在具体实施方式的上述方法中，各步骤的撰写顺序并不意味着严格的执行顺序而对实施过程构成任何限定，各步骤的具体执行顺序应当以其功能和可能的内在逻辑确定。
基于同一发明构思，本公开实施例中还提供了与交通标线的识别方法对应的交通标线的识别装置，由于本公开实施例中的装置解决问题的原理与本公开实施例上述交通标线的识别方法相似，因此装置的实施可以参见方法的实施，重复之处不再赘述。
如图3所示，为本公开实施例提供的一种交通标线的识别装置的示意图，包括：
获取模块301，用于获取道路点云数据；
划分模块302，用于对所述道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；
第一确定模块303，用于基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率；
第二确定模块304，用于基于各个所述目标栅格分别所属的交通标线类别的概率，确定所述道路点云数据的标线识别结果。
在一种可能的实施方式中，所述交通标线类别包括车道线类别和道路边界线类别；所述第一确定模块303，在所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率时，用于针对每个所述目标栅格所包含的所述局部点云数据，对所述局部点云数据中每个点云点的点云信息进行特征提取，生成所述目标栅格的道路特征信息；
基于所述道路特征信息，确定所述目标栅格属于所述车道线类别的第一概率和属于 所述道路边界线类别的第二概率。
在一种可能的实施方式中，所述第二确定模块304，在所述基于各个所述目标栅格分别所属的交通标线类别的所述概率，确定所述道路点云数据的标线识别结果时，用于针对每个所述目标栅格，确定所述目标栅格的所述第一概率和所述第二概率中是否存在大于预设阈值的目标概率；
在存在所述目标概率的情况下，根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果；
基于各个所述目标栅格的所述识别结果，确定所述道路点云数据的标线识别结果。
在一种可能的实施方式中，所述第二确定模块304，在所述根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果时，用于在所述目标概率包括所述第一概率和所述第二概率的情况下，确定所述第一概率和所述第二概率中的最大概率；
将所述最大概率关联的所述交通标线类别，作为所述目标栅格的识别结果。
在一种可能的实施方式中，所述道路点云数据为行驶装置采集的，所述装置还包括：
控制模块305，用于在所述确定所述道路点云数据的标线识别结果之后，基于所述标线识别结果指示的至少一条交通标线的标线信息，控制所述行驶装置行驶，其中，所述标线信息包括标线位置和/或标线类别。
在一种可能的实施方式中，所述第一确定模块303，在所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率时，用于利用训练后的目标神经网络，基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率。
在一种可能的实施方式中，所述装置还包括：
训练模块306，用于根据下述步骤训练得到所述目标神经网络：
获取样本点云数据；
对所述样本点云数据进行栅格化划分，得到至少一个样本栅格所包含的局部样本点云数据；并确定每个样本栅格的标注标签信息；
将每个所述样本栅格所包含的所述局部样本点云数据输入至待训练神经网络，生成所述样本栅格属于所述交通标线类别的预测概率；
基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
在一种可能的实施方式中，所述预测概率包括所述样本栅格属于车道线类别的第一预测概率、和属于道路边界线类别的第二预测概率；
所述训练模块306，在所述基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络时，用于在所述样本栅格的所述标注标签信息指示所述样本栅格不属于背景的情况下，基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失；
在所述样本栅格的所述标注标签信息指示所述样本栅格属于背景的情况下，基于所述第一预测概率、所述第二预测概率和所述背景标签对应的标注标签信息，确定第二损失；
基于所述第一损失和所述第二损失中的至少一种，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
在一种可能的实施方式中，所述第一损失包括第一子损失和第二子损失；
所述训练模块306，在所述基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失时，用于从所述样本栅格的所述第一预测概率和所述第二预测概率中，确定出与所述样本栅格的标注标签信息指示的交通标线类别相匹配的目标预测概率；
基于所述目标预测概率和所述样本栅格的标注标签信息，确定第一子损失；
基于所述第一预测概率和所述第二预测概率中除所述目标预测概率之外的其他预测概率，和背景标签对应的标注标签信息，确定第二子损失。
在一种可能的实施方式中，所述训练模块306，在所述确定每个样本栅格的标注标签信息时，用于基于各个所述样本栅格分别所包含的所述局部样本点云数据，生成样本俯视图；其中，每个所述样本栅格对应所述样本俯视图中的一个像素点；
基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
在一种可能的实施方式中，所述训练模块306，在所述基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息时，用于基于所述样本俯视图中各个像素点的像素信息，确定所述样本俯视图中每个像素点对应的标注标签信息；
针对所述标注标签信息指示为交通标线类别的目标像素点，基于预设的扩展宽度，将所述目标像素点的相邻像素点的标注标签信息调整为所述目标像素点的标注标签信息；
基于所述样本俯视图中所述相邻像素点的调整后的标注标签信息、和除所述相邻像素点之外的其他像素点的未调整的所述标注标签信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
关于装置中的各模块的处理流程、以及各模块之间的交互流程的描述可以参照上述方法实施例中的相关说明，这里不再详述。
基于同一技术构思，本申请实施例还提供了一种计算机设备。参照图4所示，为本申请实施例提供的一种计算机设备的结构示意图，包括：
处理器41、存储器42和总线43。其中，存储器42存储有处理器41可执行的机器可读指令，处理器41用于执行存储器42中存储的机器可读指令，所述机器可读指令被处理器41执行时，处理器41执行下述步骤：S101：获取道路点云数据；S102：对道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；S103：基于每个目标栅格所包含的局部点云数据，确定目标栅格属于交通标线类别的概率以及S104：基于各个目标栅格分别所属的交通标线类别的概率，确定道路点云数据的标线识别结果。
上述存储器42包括内存421和外部存储器422；这里的内存421也称内存储器，用于暂时存放处理器41中的运算数据，以及与硬盘等外部存储器422交换的数据，处理器41通过内存421与外部存储器422进行数据交换，当计算机设备运行时，处理器41与存储器42之间通过总线43通信，使得处理器41在执行上述方法实施例中所提及的执行指令。
本公开实施例还提供一种计算机可读存储介质，该计算机可读存储介质上存储有计算机程序，该计算机程序被处理器运行时执行上述方法实施例中所述的交通标线的识别方法的步骤。其中，该存储介质可以是易失性或非易失的计算机可读取存储介质。
本公开实施例所提供的交通标线的识别方法的计算机程序产品，包括存储了程序代码的计算机可读存储介质，所述程序代码包括的指令可用于执行上述方法实施例中所述的交通标线的识别方法的步骤，具体可参见上述方法实施例，在此不再赘述。
该计算机程序产品可以具体通过硬件、软件或其结合的方式实现。在一个可选实施例中，所述计算机程序产品具体体现为计算机存储介质，在另一个可选实施例中，计算机程序产品具体体现为软件产品，例如软件开发包(Software Development Kit，SDK)等等。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的装置的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。在本公开所提供的几个实施例中，应该理解到，所揭露的装置和方法，可以通过其它的方式实现。以上所描述的装置实施例仅仅是示意性的，例如，所述单元的划分，仅仅为一种逻辑功能划分，实际实现时可以有另外的划分方式，又例如，多个单元或组件可以结合，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些通信接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
所述作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个 网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本公开各个实施例中的各功能单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。
所述功能如果以软件功能单元的形式实现并作为独立的产品销售或使用时，可以存储在一个处理器可执行的非易失的计算机可读取存储介质中。基于这样的理解，本公开的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本公开各个实施例所述方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(Read-Only Memory，ROM)、随机存取存储器(Random Access Memory，RAM)、磁碟或者光盘等各种可以存储程序代码的介质。
若本申请技术方案涉及个人信息，应用本申请技术方案的产品在处理个人信息前，已明确告知个人信息处理规则，并取得个人自主同意。若本申请技术方案涉及敏感个人信息，应用本申请技术方案的产品在处理敏感个人信息前，已取得个人单独同意，并且同时满足“明示同意”的要求。例如，在摄像头等个人信息采集装置处，设置明确显著的标识告知已进入个人信息采集范围，将会对个人信息进行采集，若个人自愿进入采集范围即视为同意对其个人信息进行采集；或者在个人信息处理的装置上，利用明显的标识/信息告知个人信息处理规则的情况下，通过弹窗信息或请个人自行上传其个人信息等方式获得个人授权；其中，个人信息处理规则可包括个人信息处理者、个人信息处理目的、处理方式、处理的个人信息种类等信息。
最后应说明的是：以上所述实施例，仅为本公开的具体实施方式，用以说明本公开的技术方案，而非对其限制，本公开的保护范围并不局限于此，尽管参照前述实施例对本公开进行了详细的说明，本领域的普通技术人员应当理解：任何熟悉本技术领域的技术人员在本公开揭露的技术范围内，其依然可以对前述实施例所记载的技术方案进行修改或可轻易想到变化，或者对其中部分技术特征进行等同替换；而这些修改、变化或者替换，并不使相应技术方案的本质脱离本公开实施例技术方案的精神和范围，都应涵盖在本公开的保护范围之内。因此，本公开的保护范围应所述以权利要求的保护范围为准。

Claims (14)

  1. 一种交通标线的识别方法，其特征在于，包括：
    获取道路点云数据；
    对所述道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；
    基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率；
    基于各个所述目标栅格分别所属的交通标线类别的概率，确定所述道路点云数据的标线识别结果。
  2. 根据权利要求1所述的方法，其特征在于，所述交通标线类别包括车道线类别和道路边界线类别；所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率，包括：
    针对每个所述目标栅格所包含的所述局部点云数据，对所述局部点云数据中每个点云点的点云信息进行特征提取，生成所述目标栅格的道路特征信息；
    基于所述道路特征信息，确定所述目标栅格属于所述车道线类别的第一概率和属于所述道路边界线类别的第二概率。
  3. 根据权利要求2所述的方法，其特征在于，所述基于各个所述目标栅格分别所属的交通标线类别的概率，确定所述道路点云数据的标线识别结果，包括：
    针对每个所述目标栅格，确定所述目标栅格的所述第一概率和所述第二概率中是否存在大于预设阈值的目标概率；
    在存在所述目标概率的情况下，根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果；
    基于各个所述目标栅格的所述识别结果，确定所述道路点云数据的标线识别结果。
  4. 根据权利要求3所述的方法，其特征在于，所述根据所述目标概率关联的所述交通标线类别，确定所述目标栅格的识别结果，包括：
    在所述目标概率包括所述第一概率和所述第二概率的情况下，确定所述第一概率和所述第二概率中的最大概率；
    将所述最大概率关联的所述交通标线类别，作为所述目标栅格的识别结果。
  5. 根据权利要求1至4任一项所述的方法，其特征在于，所述道路点云数据为行驶装置采集的，在所述确定所述道路点云数据的标线识别结果之后，还包括：
    基于所述标线识别结果指示的至少一条交通标线的标线信息，控制所述行驶装置行驶，其中，所述标线信息包括标线位置和/或标线类别。
  6. 根据权利要求1至5任一项所述的方法，其特征在于，所述基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率，包括：
    利用训练后的目标神经网络，基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率。
  7. 根据权利要求6所述的方法，其特征在于，根据下述步骤训练得到所述目标神经网络：
    获取样本点云数据；
    对所述样本点云数据进行栅格化划分，得到至少一个样本栅格所包含的局部样本点云数据；并确定每个样本栅格的标注标签信息；
    将每个所述样本栅格所包含的所述局部样本点云数据输入至待训练神经网络，生成所述样本栅格属于所述交通标线类别的预测概率；
    基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
  8. 根据权利要求7所述的方法，其特征在于，所述预测概率包括所述样本栅格属于车道线类别的第一预测概率、和属于道路边界线类别的第二预测概率；
    所述基于各个所述样本栅格分别所属的交通标线类别的预测概率、和每个所述样本栅格的标注标签信息，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络，包括：
    在所述样本栅格的所述标注标签信息指示所述样本栅格不属于背景的情况下，基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失；
    在所述样本栅格的所述标注标签信息指示所述样本栅格属于背景的情况下，基于所述第一预测概率、所述第二预测概率和所述背景标签对应的标注标签信息，确定第二损失；
    基于所述第一损失和所述第二损失中的至少一种，对所述待训练神经网络进行迭代训练，直至满足训练截止条件，得到所述目标神经网络。
  9. 根据权利要求8所述的方法，其特征在于，所述第一损失包括第一子损失和第 二子损失；
    所述基于所述样本栅格的所述第一预测概率、所述第二预测概率、所述样本栅格的标注标签信息、和背景标签对应的标注标签信息，确定第一损失，包括：
    从所述样本栅格的所述第一预测概率和所述第二预测概率中，确定出与所述样本栅格的标注标签信息指示的交通标线类别相匹配的目标预测概率；
    基于所述目标预测概率和所述样本栅格的标注标签信息，确定第一子损失；
    基于所述第一预测概率和所述第二预测概率中除所述目标预测概率之外的其他预测概率，和背景标签对应的标注标签信息，确定第二子损失。
  10. 根据权利要求7至9任一项所述的方法，其特征在于，所述确定每个样本栅格的标注标签信息，包括：
    基于各个所述样本栅格分别所包含的所述局部样本点云数据，生成样本俯视图；其中，每个所述样本栅格对应所述样本俯视图中的一个像素点；
    基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
  11. 根据权利要求10所述的方法，其特征在于，所述基于所述样本俯视图中各个像素点的像素信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息，包括：
    基于所述样本俯视图中各个像素点的像素信息，确定所述样本俯视图中每个像素点对应的标注标签信息；
    针对所述标注标签信息指示为交通标线类别的目标像素点，基于预设的扩展宽度，将所述目标像素点的相邻像素点的标注标签信息调整为所述目标像素点的标注标签信息；
    基于所述样本俯视图中所述相邻像素点的调整后的标注标签信息、和除所述相邻像素点之外的其他像素点的未调整的所述标注标签信息，确定与每个所述像素点匹配的所述样本栅格的标注标签信息。
  12. 一种交通标线的识别装置，其特征在于，包括：
    获取模块，用于获取道路点云数据；
    划分模块，用于对所述道路点云数据进行栅格化划分，得到至少一个目标栅格所包含的局部点云数据；
    第一确定模块，用于基于每个所述目标栅格所包含的所述局部点云数据，确定所述目标栅格属于交通标线类别的概率；
    第二确定模块，用于基于各个所述目标栅格分别所属的交通标线类别的概率，确定所述道路点云数据的标线识别结果。
  13. 一种计算机设备，其特征在于，包括：处理器、存储器，所述存储器存储有所述处理器可执行的机器可读指令，所述处理器用于执行所述存储器中存储的机器可读指令，所述机器可读指令被所述处理器执行时，所述处理器执行如权利要求1至11任意一项所述的交通标线的识别方法的步骤。
  14. 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上存储有计算机程序，所述计算机程序被计算机设备运行时，所述计算机设备执行如权利要求1至11任意一项所述的交通标线的识别方法的步骤。
PCT/CN2023/102444 2022-07-01 2023-06-26 交通标线的识别方法、装置、计算机设备和存储介质 Ceased WO2024002014A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202210773244.8 2022-07-01
CN202210773244.8A CN115131759B (zh) 2022-07-01 2022-07-01 交通标线的识别方法、装置、计算机设备和存储介质

Publications (1)

Publication Number Publication Date
WO2024002014A1 true WO2024002014A1 (zh) 2024-01-04

Family

ID=83381858

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2023/102444 Ceased WO2024002014A1 (zh) 2022-07-01 2023-06-26 交通标线的识别方法、装置、计算机设备和存储介质

Country Status (2)

Country Link
CN (1) CN115131759B (zh)
WO (1) WO2024002014A1 (zh)

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118999600A (zh) * 2024-10-24 2024-11-22 北京集度科技有限公司 一种道路标线信息确定方法、车辆、设备及产品
CN119091410A (zh) * 2024-11-04 2024-12-06 北京集度科技有限公司 一种道路标线检测方法、车辆、设备及产品
CN121545132A (zh) * 2026-01-21 2026-02-17 立得空间信息技术股份有限公司 基于车载点云图像的道路要素识别方法、装置及存储介质

Families Citing this family (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115131759B (zh) * 2022-07-01 2025-06-06 上海商汤临港智能科技有限公司 交通标线的识别方法、装置、计算机设备和存储介质
CN116503383B (zh) * 2023-06-20 2023-09-12 上海主线科技有限公司 道路曲线检测方法、系统及介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363054A (zh) * 2018-11-16 2019-10-22 北京京东尚科信息技术有限公司 道路标识线识别方法、装置和系统
WO2020103892A1 (zh) * 2018-11-21 2020-05-28 北京市商汤科技开发有限公司 车道线检测方法、装置、电子设备及可读存储介质
CN113298910A (zh) * 2021-05-14 2021-08-24 阿波罗智能技术(北京)有限公司 生成交通标志线地图的方法、设备和存储介质
CN113971221A (zh) * 2020-07-22 2022-01-25 上海商汤临港智能科技有限公司 一种点云数据的处理方法、装置、电子设备及存储介质
CN115131759A (zh) * 2022-07-01 2022-09-30 上海商汤临港智能科技有限公司 交通标线的识别方法、装置、计算机设备和存储介质

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108216229B (zh) * 2017-09-08 2020-01-10 北京市商汤科技开发有限公司 交通工具、道路线检测和驾驶控制方法及装置
US10657390B2 (en) * 2017-11-27 2020-05-19 Tusimple, Inc. System and method for large-scale lane marking detection using multimodal sensor data
KR102633140B1 (ko) * 2018-10-23 2024-02-05 삼성전자주식회사 주행 정보를 결정하는 방법 및 장치
CN111209780A (zh) * 2018-11-21 2020-05-29 北京市商汤科技开发有限公司 车道线属性检测方法、装置、电子设备及可读存储介质
CN111353512B (zh) * 2018-12-20 2023-07-28 长沙智能驾驶研究院有限公司 障碍物分类方法、装置、存储介质和计算机设备
KR20210102182A (ko) * 2020-02-07 2021-08-19 선전 센스타임 테크놀로지 컴퍼니 리미티드 도로 표시 인식 방법, 지도 생성 방법, 및 관련 제품
CN113970758A (zh) * 2020-07-22 2022-01-25 上海商汤临港智能科技有限公司 点云数据处理的方法、及装置
CN113191261B (zh) * 2021-04-29 2022-12-06 北京百度网讯科技有限公司 图像类别的识别方法、装置和电子设备

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN110363054A (zh) * 2018-11-16 2019-10-22 北京京东尚科信息技术有限公司 道路标识线识别方法、装置和系统
WO2020103892A1 (zh) * 2018-11-21 2020-05-28 北京市商汤科技开发有限公司 车道线检测方法、装置、电子设备及可读存储介质
CN113971221A (zh) * 2020-07-22 2022-01-25 上海商汤临港智能科技有限公司 一种点云数据的处理方法、装置、电子设备及存储介质
CN113298910A (zh) * 2021-05-14 2021-08-24 阿波罗智能技术(北京)有限公司 生成交通标志线地图的方法、设备和存储介质
CN115131759A (zh) * 2022-07-01 2022-09-30 上海商汤临港智能科技有限公司 交通标线的识别方法、装置、计算机设备和存储介质

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN118999600A (zh) * 2024-10-24 2024-11-22 北京集度科技有限公司 一种道路标线信息确定方法、车辆、设备及产品
CN119091410A (zh) * 2024-11-04 2024-12-06 北京集度科技有限公司 一种道路标线检测方法、车辆、设备及产品
CN121545132A (zh) * 2026-01-21 2026-02-17 立得空间信息技术股份有限公司 基于车载点云图像的道路要素识别方法、装置及存储介质

Also Published As

Publication number Publication date
CN115131759B (zh) 2025-06-06
CN115131759A (zh) 2022-09-30

Similar Documents

Publication Publication Date Title
Min et al. Traffic sign recognition based on semantic scene understanding and structural traffic sign location
US12585944B2 (en) Data processing system, object detection method, and apparatus thereof
CN110148196B (zh) 一种图像处理方法、装置以及相关设备
Hoang et al. Enhanced detection and recognition of road markings based on adaptive region of interest and deep learning
CN111709923B (zh) 一种三维物体检测方法、装置、计算机设备和存储介质
US11410549B2 (en) Method, device, readable medium and electronic device for identifying traffic light signal
CN115131759B (zh) 交通标线的识别方法、装置、计算机设备和存储介质
CN108280397B (zh) 基于深度卷积神经网络的人体图像头发检测方法
CN113674287A (zh) 高精地图的绘制方法、装置、设备以及存储介质
CN108564088A (zh) 车牌识别方法、装置、设备及可读存储介质
CN101859382A (zh) 一种基于最大稳定极值区域的车牌检测与识别的方法
CN111259796A (zh) 一种基于图像几何特征的车道线检测方法
CN112819753B (zh) 一种建筑物变化检测方法、装置、智能终端及存储介质
CN115631344A (zh) 一种基于特征自适应聚合的目标检测方法
Mijić et al. Traffic sign detection using YOLOv3
CN115063765B (zh) 一种道路边界线确定方法、装置、设备及存储介质
CN114399768A (zh) 基于Tesseract-OCR引擎的工件产品序列号识别方法、装置及系统
CN116129386A (zh) 可行驶区域检测方法、系统及计算机可读介质
CN113449629B (zh) 基于行车视频的车道线虚实识别装置、方法、设备及介质
Qiu et al. Lightweight cross-modal information measure and propagation for road extraction from remote sensing image and trajectory/LiDAR
CN115760878B (zh) 三维图像实体分割方法、装置、设备、存储介质及车辆
Daraghmi et al. Accurate real-time traffic sign recognition based on the connected component labeling and the color histogram algorithms
CN106529391B (zh) 一种鲁棒的限速交通标志检测与识别方法
CN114565902A (zh) 基于联通区域提取和关键点拟合的车道检测方法及系统
CN119380311B (zh) 一种交通路牌目标检测识别方法和系统

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: 23830191

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: 23830191

Country of ref document: EP

Kind code of ref document: A1

32PN Ep: public notification in the ep bulletin as address of the adressee cannot be established

Free format text: NOTING OF LOSS OF RIGHTS PURSUANT TO RULE 112(1) EPC (EPO FORM 1205A DATED 11/06/2025)

122 Ep: pct application non-entry in european phase

Ref document number: 23830191

Country of ref document: EP

Kind code of ref document: A1