CN112347983B

CN112347983B - Lane line detection processing method, lane line detection processing device, computer equipment and storage medium

Info

Publication number: CN112347983B
Application number: CN202011362122.7A
Authority: CN
Inventors: 刘春�
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2020-11-27
Filing date: 2020-11-27
Publication date: 2021-12-14
Anticipated expiration: 2040-11-27
Also published as: CN112347983A

Abstract

The application relates to a lane line detection processing method, a lane line detection processing device, computer equipment and a storage medium. The method comprises the following steps: obtaining a plurality of candidate lane lines corresponding to a target traffic area; selecting at least two lane lines from the candidate lane lines to obtain an initial lane line combination; acquiring a spatial distribution relation among lane lines in the initial lane line combination, and acquiring lane line distribution scores corresponding to the spatial distribution relation; determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score; and optimizing the initial lane line combination based on the first combination detection score to obtain a target lane line combination. The method can improve the accuracy of the lane line. The lane line in the method can be detected based on an artificial intelligent lane line detection model, and the method is applied to the field of automatic driving and can improve the safety of automatic driving vehicles.

Description

Lane line detection processing method, lane line detection processing device, computer equipment and storage medium

Technical Field

The present application relates to the field of information processing technologies, and in particular, to a lane line detection processing method and apparatus, a computer device, and a storage medium.

Background

As roads are built, the number of lanes increases, the relationship between lanes is complicated, and in order to grasp more information on the roads, users often select a high-precision map to assist driving. The high-precision map is an indispensable data base for improving the vehicle automation and driving safety level in advanced auxiliary driving and automatic driving, and is used for realizing the functions of vehicle self-positioning and remote environment perception.

In the process of producing a high-precision map, it is usually necessary to detect a lane line, for example, the lane line may be detected based on an artificial intelligent lane line detection model, but the accuracy of the lane line detected by the current lane line detection processing method is low.

Disclosure of Invention

In view of the above, it is necessary to provide a lane line detection processing method, apparatus, computer device, and storage medium capable of improving lane line accuracy.

A lane line detection processing method, the method comprising: obtaining a plurality of candidate lane lines corresponding to a target traffic area; selecting at least two lane lines from the candidate lane lines to obtain an initial lane line combination; acquiring a spatial distribution relation among lane lines in the initial lane line combination, and acquiring lane line distribution scores corresponding to the spatial distribution relation; determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score; and optimizing the initial lane line combination based on the first combination detection score to obtain a target lane line combination.

A lane line detection processing apparatus, the apparatus comprising: the candidate lane line acquisition module is used for acquiring a plurality of candidate lane lines corresponding to the target traffic area; an initial lane line combination obtaining module, configured to select at least two lane lines from the multiple candidate lane lines to obtain an initial lane line combination; the lane line distribution score acquisition module is used for acquiring a spatial distribution relationship among lane lines in the initial lane line combination and acquiring lane line distribution scores corresponding to the spatial distribution relationship; the first combination detection score determining module is used for determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score; and the target lane combination obtaining module is used for optimizing the initial lane combination based on the first combination detection score to obtain a target lane combination.

In some embodiments, the lane line distribution score acquisition module includes: a spatial distribution relation obtaining unit, configured to obtain a spatial distribution relation between lane lines in the initial lane line combination; a distribution relation quantity obtaining unit, configured to perform classification statistics on the spatial distribution relations according to distribution relation types to obtain distribution relation quantities corresponding to the distribution relation types; and the lane line distribution score determining unit is used for determining the lane line distribution score according to the distribution relation quantity corresponding to the distribution relation type.

In some embodiments, the lane line distribution score obtaining unit is further configured to obtain a relationship weight corresponding to the distribution relationship type; and carrying out weighted summation according to the relation weight and the distribution relation quantity corresponding to the distribution relation type to obtain the lane line distribution score.

In some embodiments, the lane line distribution score obtaining unit is further configured to obtain a negative cost score according to the number of distribution relations corresponding to the distribution relation types when the spatial distribution relation corresponding to the distribution relation type satisfies the lane line distribution relation; the distribution relation quantity corresponding to the distribution relation type and the negative cost score form a negative correlation relation; and obtaining the lane line distribution score according to the negative cost score.

In some embodiments, the lane line distribution score obtaining unit is further configured to obtain a forward cost score according to the number of distribution relations corresponding to the distribution relation types when the spatial distribution relation corresponding to the distribution relation types deviates from the lane line distribution relation; wherein, the distribution relation quantity corresponding to the distribution relation type and the forward cost score form a positive correlation; and obtaining the lane line distribution score according to the forward cost score.

In some embodiments, the first combined detection score determination module comprises: a lane attribute feature obtaining unit, configured to obtain lane attribute features corresponding to each initial lane line in the initial lane line combination; the lane line attribute score acquiring unit is used for acquiring a lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics; and the first combination detection score obtaining unit is used for determining a first combination detection score corresponding to the initial lane line combination according to the lane line attribute score and the lane line distribution score.

In some embodiments, the lane line attribute characteristics include lane line length and lane line angle; the lane line attribute score is a cost score, and the lane line attribute score obtaining unit is further configured to obtain a length cost score corresponding to the initial lane line according to the length of the lane line; the length cost fraction is in a negative correlation with the lane line length; obtaining an angle cost score corresponding to the initial lane line according to the lane line angle; the angle cost fraction and the lane line angle form a positive correlation; and selecting the score with the maximum substitution value from the length cost score corresponding to the initial lane line and the angle cost score corresponding to the initial lane line as the attribute score of the lane line corresponding to the initial lane line.

In some embodiments, the target lane line combination derivation module comprises: a second combination detection score obtaining unit, configured to obtain a current lane line from the initial lane line combination, and obtain a second combination detection score obtained by subtracting the current lane line from the initial lane line combination; a score change value acquisition unit configured to acquire a score change value of the second combination detection score with respect to the first combination detection score; and the target lane line combination obtaining unit is used for processing the current lane line in the initial lane line combination according to the fractional change value to obtain the target lane line combination.

In some embodiments, the target lane line combination obtaining unit is further configured to process a current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination; selecting a lane line from the plurality of candidate lane lines, and adding the lane line into the processed lane line combination to obtain an initial lane line combination of the next round; and returning to the steps of obtaining the spatial distribution relation among the lane lines in the initial lane line combination and obtaining the lane line distribution fraction corresponding to the spatial distribution relation until a stop condition is met, and obtaining a target lane line combination.

In some embodiments, the score variation value is a cost variation value, and the target lane line combination obtaining unit is further configured to obtain an acceptance probability corresponding to the current lane line when the cost variation value is greater than a first cost threshold; determining the optimization processing operation corresponding to the current lane line from the initial lane line combination according to the receiving probability; and processing the current lane line in the initial lane line combination according to the optimization processing operation to obtain a processed lane line combination.

In some embodiments, the score variation value is a cost variation value, and the target lane line combination obtaining unit is further configured to delete the current lane line from the initial lane line combination to obtain the processed lane line combination when the cost variation value is smaller than a first cost threshold.

A computer device includes a memory in which a computer program is stored and a processor that implements the steps of the lane line detection processing method described above when executing the computer program.

A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, realizes the steps of the lane line detection processing method described above.

The lane line detection processing method, the lane line detection processing device, the computer equipment and the storage medium acquire a plurality of candidate lane lines corresponding to a target traffic area, select at least two lane lines from the candidate lane lines to obtain an initial lane line combination, acquire a spatial distribution relationship among the lane lines in the initial lane line combination, acquire lane line distribution scores corresponding to the spatial distribution relationship, determine a first combination detection score corresponding to the initial lane line combination according to the lane line distribution scores, and optimize the initial lane line combination based on the first combination detection score to obtain the target lane line combination. The first combination detection score is determined according to the lane line distribution score, and the lane line distribution score is obtained according to the spatial distribution relation among the lane lines, so that the first combination detection score can reflect the spatial distribution relation among the lane lines in the combination, the accuracy of the first combination detection score is improved, the initial lane line combination is optimized based on the first combination detection score, the target lane line combination is obtained, and the accuracy of the target lane line combination is improved.

Drawings

FIG. 1 is a diagram of an exemplary embodiment of a lane marking detection process;

FIG. 2 is a schematic flow chart of a lane line detection processing method in some embodiments;

FIG. 3 is a schematic view of the orientation angle in some embodiments;

FIG. 4 is a graph comparing before lane line detection and after lane line detection in some embodiments;

FIG. 5 is a graph of cost scores as a function of iteration number in some embodiments;

FIG. 6 is a graph of the number of lane lines as a function of iteration number in some embodiments;

FIG. 7 is a graph of increasing and decreasing number of lane lines as a function of iteration number in some embodiments;

FIG. 8 is a schematic flow chart of a lane line detection processing method in some embodiments;

FIG. 9 is a block diagram of a lane line detection processing apparatus in some embodiments;

FIG. 10 is a diagram of the internal structure of a computer device in some embodiments.

Detailed Description

In order to make the objects, technical solutions and advantages of the present application more apparent, the present application is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the present application and are not intended to limit the present application.

The lane line detection processing method provided by the application can be applied to the application environment shown in fig. 1. The mobile acquisition terminal 102 may transmit the acquired image data of the traffic area or the point cloud data of the traffic area to the first server 104, the first server 104 may perform lane line identification on the image data or the point cloud data by using a deep learning algorithm, and transmit the lane line obtained by the identification to the second server 106, and the second server 106 may optimize the lane line obtained by the identification to obtain an optimized lane line, for example, select a lane line satisfying a spatial distribution relationship of the lane line from the lane lines obtained by the identification as the optimized lane line. The mobile acquisition terminal 102, the first server 104, and the second server 106 communicate over a network. The second server 106 obtains a plurality of candidate lane lines corresponding to the target traffic area, where the candidate lane lines may be, for example, lane lines identified by the first server 104. The second server 106 may select at least two lane lines from the multiple candidate lane lines to obtain an initial lane line combination, obtain a spatial distribution relationship between the lane lines in the initial lane line combination, obtain lane line distribution scores corresponding to the spatial distribution relationship, determine a first combination detection score corresponding to the initial lane line combination according to the lane line distribution scores, and perform optimization processing on the initial lane line combination based on the first combination detection score to obtain a target lane line combination.

It is to be understood that the above application scenario is only an example, and does not constitute a limitation to the method provided in the embodiment of the present application, and the method provided in the embodiment of the present application may also be applied in other scenarios, for example, the mobile acquisition terminal 102 may perform the steps of lane line identification and lane line optimization, obtain an optimized lane line, upload the optimized lane line to the server, or update the lane line in the local map by using the optimized lane line, for example, update the lane line in the local high-precision map. Specifically, the mobile collection terminal 102 may perform lane recognition on the collected image data or point cloud data by using a deep learning algorithm, and optimize the recognized lane line by using the lane line detection processing method provided by the present application to obtain an optimized lane line. The first server and the second server of the embodiment of the present application may also be the same server. In the latter case, the terminal may cooperate with the server to perform the steps of lane line identification and lane line detection.

The resulting combination of target lane lines can be used to generate a High precision Map (HD Map). The High precision Map (HD Map) is an indispensable data base for improving the vehicle automation and driving safety level in advanced assistant driving and automatic driving, and is used for realizing the functions of vehicle self-positioning and remote environment perception. The high-precision map can also be called a high-precision map, and can be a lane-level map with the precision of centimeter level. The high-precision map can comprise lane lines, pedestrian crossings, traffic lights and speed-limiting signs.

The mobile collection terminal 102 may be a mobile terminal carrying an image collection device or a point cloud collection device, and the mobile terminal may be a vehicle collecting ground traffic data, for example. The image capturing device may be any device capable of capturing an image, for example, a camera. The point cloud collecting device may be any device capable of collecting a point cloud, and may be a laser radar, for example. But not limited to, various personal computers, notebook computers, smart phones, tablet computers, vehicle-mounted systems, and portable wearable devices, the first server 104 and the second server 106 may be independent physical servers, may also be a server cluster or distributed system formed by a plurality of physical servers, and may also be cloud servers providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, Network services, cloud communication, middleware services, domain name services, security services, a CDN (Content Delivery Network), and big data and artificial intelligence platforms.

In the present application, a model based on Artificial Intelligence (AI) is used to detect lane lines, which is a theory, method, technique and application system that simulates, extends and expands human Intelligence using a digital computer or a machine controlled by a digital computer, senses the environment, acquires knowledge and uses the knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

With the research and progress of artificial intelligence technology, the artificial intelligence technology is developed and applied in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, robots, smart medical care, smart customer service, and the like.

The scheme provided by the embodiment of the application relates to the technologies of automatic driving of artificial intelligence and the like. The automatic driving technology generally comprises technologies such as high-precision maps, environment perception, behavior decision, path planning, motion control and the like, and the self-determined driving technology has wide application prospects. The following examples are intended to illustrate in particular: the second server 106 may transmit the generated high-precision map to the autonomous vehicle in generating the high-precision map using the obtained target lane line combination, and the autonomous vehicle may plan a movement route according to the lane lines shown in the high-precision map, thereby improving safety of autonomous driving.

In some embodiments, as shown in fig. 2, a lane line detection processing method is provided, which is described by taking the method as an example applied to the second server 106 in fig. 1, and includes the following steps:

s202, a plurality of candidate lane lines corresponding to the target traffic area are obtained.

The traffic area refers to an area including a road on which the vehicle travels. The target traffic zone may be any traffic zone. The lane line candidates belong to the same target traffic area, and for example, an area corresponding to a preset position range may be used as the target traffic area. The target traffic area may be image-captured, lane lines may be extracted from the captured image, and each lane line obtained by the extraction may be used as a candidate lane line.

The lanes are obtained by artificially dividing the roads by lane lines, namely, different lanes are distinguished by lane lines. A road may be divided into lanes. The autonomous vehicle can travel along a lane by recognizing a lane line in a road while traveling. The lane line refers to a boundary line of the lane, and may include at least one of a left boundary line or a right boundary line. Fig. 3 shows a lane line in a road.

The lane lines may be two-dimensional. The lane lines may be represented in the form of polylines (polylines). The lane lines may be stored in the form of polygonal lines, which are stored by storing respective shape point coordinates of the polygonal lines. One lane line may be formed by a plurality of polygonal lines, the polygonal lines are formed by connecting shape points, and the plurality of polygonal lines connected end to end may describe one lane line. Each lane line may be represented by a corresponding set of shape points. The set of shape points includes a plurality of shape points. Each shape point corresponds to shape point location information. Shape point location information refers to the location of a shape point and may include longitude and latitude coordinates.

The candidate lane lines may be extracted from the image or the point cloud, for example, by using a deep learning algorithm. By merely identifying the lane lines in the image, the accuracy of the obtained lane lines is low, i.e., there may be erroneously identified lane lines or redundant lane lines among a plurality of candidate lane lines. Therefore, the plurality of candidate lane lines can be optimized, that is, non-real lane lines or redundant lane lines are filtered out from the plurality of candidate lane lines.

Specifically, the second server may obtain a plurality of candidate lane lines corresponding to the target traffic area when receiving the lane line detection request, for example, when the second server receives the lane line detection request sent by the first server, the lane line detection request may carry each lane line to be detected, and the second server may respond to the lane line detection request to detect each lane line carried in the lane line detection request. The second server may also acquire a plurality of candidate lane lines corresponding to the target traffic area according to a specific time interval, where the specific time interval may be preset, for example, 24 hours, or may be set according to actual needs. Of course, the second server may also periodically detect whether the lane line corresponding to the target traffic area in the first server is updated, and when it is determined that the lane line corresponding to the target traffic area in the first server is updated, obtain the updated lane line as the candidate lane line.

In some embodiments, the first server may store candidate lane lines corresponding to a plurality of traffic areas, where different candidate lane lines may be obtained by identifying images or point clouds acquired at different time periods, for example, the candidate lane line 1 is obtained by identifying an image acquired on a first day, and the candidate lane line 2 is obtained by identifying an image acquired on a second day. The second server may obtain the lane line corresponding to the same time period from the first server as a candidate lane line.

In some embodiments, the second server may obtain lane lines of different time periods corresponding to the target traffic area, and obtain a plurality of candidate lane lines corresponding to different time periods respectively. Respectively detecting a plurality of candidate lane lines corresponding to different time periods to obtain target lane line combinations corresponding to different time periods, calculating the same lane line in the target lane line combinations corresponding to different time periods to form an optimized lane line combination, and generating a high-precision map by using the optimized lane line combination.

S204, selecting at least two lane lines from the candidate lane lines to obtain an initial lane line combination.

Specifically, the initial lane line combination includes a plurality of initial lane lines, a plurality referring to at least two. The initial lane line is selected from a plurality of candidate lane lines. At least two lane lines can be randomly selected from the plurality of candidate lane lines to obtain an initial lane line combination. Or selecting at least two lane lines from a plurality of candidate lane lines according to the attribute characteristics of the lane lines to obtain an initial lane line combination. The second server may sort lane lines of the plurality of candidate lane lines according to the lane line attribute characteristics, select a lane line satisfying a lane line selection condition from the plurality of candidate lane lines as an initial lane line, and form an initial lane line combination. The lane line selection condition may include at least one of a ranking before a preset ranking or a preset ratio before. The preset ordering is preset and may be, for example, 200. The preset ratio is a preset ratio, and may be a ratio between the preset sequence and the maximum sequence number, for example, the preset sequence is 200, and the maximum sequence number is 1000, the preset ratio is 1/5.

The lane line attribute feature refers to a feature of a lane line, and may include any feature of the lane line, for example, may include at least one of a lane line length or a lane line angle. The lane line length refers to the length of the lane line. The lane line angle may be used to reflect the degree of curvature or smoothness of the shape of the lane line, and the larger the lane line angle is, the more curved the lane line is, or the less smooth the lane line is, i.e. the larger the difference between the lane line and the straight line is, the smaller the lane line angle is, the smoother the lane line is, or the smaller the difference between the lane line and the straight line is. The lane line angle may be calculated from the direction angles of the shape points corresponding to the lane line, and for example, a statistical value of the direction angles corresponding to the respective shape points may be used as the lane line angle. The statistical value of the orientation angles of the respective shape points may be any one of a mean value and a variance of the orientation angles of the respective shape points. The lane line angle may also be an included angle between the direction of the lane line and a specific direction, and the specific direction may be preset or may be set according to needs, for example, may be a 90-degree direction.

The direction angle corresponding to the shape point may be calculated from the direction angles corresponding to the direction lines passing through the shape point and the shape points adjacent to the shape point. When the first shape point is included in the set of shape points, the shape point adjacent to the first shape point may be a shape point of the set of shape points whose distance from the first shape point is smallest. For example, as shown in fig. 3, the shape point set includes shape point 1, shape point 2, and shape point 3, where shape point 1 is a forward adjacent shape point of shape point 2, and shape point 3 is a backward adjacent shape point of shape point 2. The first direction line is a direction line passing through the shape point 1 and the shape point 2 and having a direction from the shape point 1 to the shape point 2. The second direction line is a direction line passing through the shape point 2 and the shape point 3 and having a direction from the shape point 2 to the shape point 3. The first direction angle corresponding to the first direction line may be an included angle between the first direction line and a horizontal line, and the second direction angle corresponding to the second direction line may be an included angle between the second direction line and the horizontal line. The direction angle corresponding to the shape point 2 may be calculated from the second direction angle and the first direction angle, and for example, an average value of the second direction angle and the first direction angle may be used as the direction angle corresponding to the shape point 2.

In some embodiments, when the lane lines of the plurality of candidate lane lines are sorted according to the lane line attribute feature, the lane lines of the plurality of candidate lane lines may be sorted according to the lane line length feature, for example, the lane line length may be sorted from large to small, or the lane lines of the plurality of candidate lane lines may be sorted according to the lane line shape feature, for example, the lane line curvature may be sorted from small to large. The second server may select, from the plurality of candidate lane lines, each lane line ranked before a preset ranking to obtain an initial lane line combination. The larger the length of the lane line is, the higher the probability of being a real lane line is, and the smaller the degree of curvature of the lane line is, the higher the probability of being a real lane line is. Therefore, more real lane lines can be obtained by selecting each lane line which is sequenced before the preset sequencing. Thereby accelerating the convergence speed of the optimization algorithm.

And S206, acquiring the spatial distribution relationship among the lane lines in the initial lane line combination, and acquiring lane line distribution scores corresponding to the spatial distribution relationship.

The spatial distribution relationship between the lane lines refers to a distribution relationship between the lane lines in space, and may include at least one of a "cross" relationship, an "overlap" relationship, or a "connection" relationship. The "connection" relationship means that the tail of the lane line coincides with the head of the lane line, "overlap" relationship means that a part or all of the regions between two or more lane lines coincide, "cross" relationship means that two or more lane lines pass through each other. The control distribution relationship between lane lines can be represented by a Stochastic Geometry model (Stochastic Geometry). The stochastic geometry model is a mathematical model for researching a pattern (pattern) of geometric objects arranged in space, is suitable for researching the spatial distribution of the geometric objects in the natural and artificial environments, and can research the average statistical property by abstracting nodes of the spatial distribution into certain probability distribution.

The spatial distribution relationship between the lane lines may be a distribution relationship of two lane lines in space, or a distribution relationship of more than two lane lines in space, for example, a distribution relationship of three lane lines in space.

The lane line distribution score may be used to reflect the distribution between lane lines in the initial lane line combination. The lane line distribution score may be a cost score or a distribution profit score. When the cost score is obtained, the smaller the lane line distribution score is, the more the distribution of the lane lines in the initial lane line combination satisfies the distribution rule of the lane lines, and the larger the lane line distribution score is, the more the distribution of the lane lines in the initial lane line combination deviates from the distribution rule of the lane lines. When the profit score is obtained, the smaller the lane line distribution score is, the more the distribution of the lane lines in the initial lane line combination deviates from the distribution rule of the lane lines, and the larger the lane line distribution score is, the more the distribution of the lane lines in the initial lane line combination satisfies the distribution rule of the lane lines.

The cost score is used for representing the difference between the actual result and the ideal result, and the smaller the cost score is, the smaller the difference between the actual result and the ideal result is represented, and the larger the cost score is, the larger the difference between the actual result and the ideal result is represented. The profit score may also be used to represent the difference between the actual result and the ideal result, with a smaller profit score representing a greater difference between the actual result and the ideal result and a larger profit score representing a lesser difference between the actual result and the ideal result. The actual result may be, for example, a lane line distribution relationship between lane lines in the initial lane line combination, and the ideal result may be, for example, a standard lane line distribution relationship, i.e., a distribution relationship satisfying a lane line distribution rule, such as a "connection" relationship.

Specifically, the correspondence between the spatial distribution relationship and the lane line distribution score is preset, so that the corresponding lane line distribution score can be obtained according to the spatial distribution relationship. There may be a plurality of spatial distribution relations between lane lines in the initial lane line combination, and there may also be a plurality of the same spatial distribution relation, for example, there are 3 pairs of lane lines, and the spatial distribution relations between the 3 pairs of lane lines are all "overlapping" relations. The second server may obtain lane line distribution scores corresponding to the same spatial distribution relationship. For example, the lane line distribution score may be calculated according to the number of lane line pairs corresponding to the same spatial distribution relationship.

In some embodiments, the spatial distribution relationship may be divided into a spatial distribution relationship satisfying the lane line distribution relationship and a spatial distribution relationship deviating from the lane line distribution relationship according to the lane line distribution relationship. The lane line distribution relationship refers to a normal lane line distribution relationship, and may include, for example, a "connection" relationship. The spatial distribution relationship satisfying the distribution relationship of the lane lines refers to a spatial distribution relationship belonging to the distribution relationship of the lane lines. The spatial distribution relationship deviating from the distribution relationship of the lane lines refers to a spatial distribution relationship that does not belong to the distribution relationship of the lane lines. For example, the spatial distribution relationship satisfying the lane line distribution relationship may include "connection", for example. The spatial distribution relationship deviating from the lane line distribution relationship may include, for example, "overlap" and "cross". The second server may calculate a score corresponding to a spatial distribution relationship satisfying the lane line distribution relationship and a score corresponding to a spatial distribution relationship deviating from the lane line distribution relationship, and take the sum of the two scores as the lane line distribution score.

S208, determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score.

The first combined detection score and the lane line distribution score form a positive correlation relationship, the larger the lane line distribution score is, the larger the first combined detection score is, and the smaller the lane line distribution score is, the smaller the first combined detection score is.

Specifically, the second server may perform statistical calculation on the distribution scores of the lane lines, and use the calculated result as the first combined detection score. The statistical calculation may be, for example, any one of summation and mean calculation.

In some embodiments, the second server may obtain lane line attribute features corresponding to each initial lane line in the initial lane line combinations, obtain lane line attribute scores corresponding to the lane line attribute features, and determine first combination detection scores corresponding to the initial lane line combinations according to the lane line attribute scores and the lane line distribution scores. For example, the second server may calculate an attribute score statistic for each lane line attribute score, and the attribute score statistic may be, for example, a result obtained by summing up the lane line attribute scores or a result obtained by performing an average calculation. The second server may calculate the first combined detection score according to the attribute score statistic and the lane line distribution score, for example, a result obtained by adding the attribute score statistic and the lane line distribution score may be used as the first combined detection score. The first combined detection score may be calculated, for example, using equation (1). U ═ Udata + Uprior (1). Wherein, U represents the first combined detection score, Udata represents the attribute score statistic, which may also be referred to as a data item, and Uprior represents the lane line distribution score, which may also be referred to as a prior item. Where Udata ═ Σ Udata. udata represents the lane line attribute score.

S210, optimizing the initial lane line combination based on the first combination detection score to obtain a target lane line combination.

Wherein the optimization process may be at least one of increasing lane lines or decreasing lane lines. The target lane line combination is obtained by optimizing the initial lane line combination. The initial lane line combination may be optimized by various optimization methods, may be a custom optimization algorithm, or may be an existing optimization algorithm, for example, any one of a Multiple Birth and Death algorithm (MBD) genetic algorithm or an ant colony algorithm may be used. The multi-birth-death algorithm is one of markov chain monte carlo (Reversible Jump MCMC, RJMCMC) that can Jump. Markov Chain Monte Carlo (MCMC) is a general name of a group of random optimization methods in statistics, is a method for sampling from complex probability distribution, and can be used for optimization of high-dimensional complex objective functions or cost functions. When the objective function and the objective model to be optimized relate to the change of parameter dimension or object dimension, and jump between different dimensions is needed, a jumpable Markov chain Monte Carlo can be adopted. In the conventional RJMCMC, one geometric object is added (dirty) or one object is reduced (dirty) in each step, and the efficiency is low. In the multi-life and death algorithm, a plurality of objects are added and reduced at the same time in each step, and the convergence rate is higher than that of the common RJMCMC.

Specifically, the second server may add a new lane line or delete an existing lane line in the initial lane line combination such that the lane line distribution in the added lane line or the deleted initial lane line combination progresses toward a better direction, for example, when the combination detection score is the cost score, the initial lane line combination may be processed toward a smaller direction of the combination detection score, and when the combination detection score is the profit score, the initial lane line combination may be processed toward a larger direction of the combination detection score. As shown in fig. 4, the plurality of candidate lane lines may be, for example, each lane line in (a) in fig. 4, where lane line 1 is an extra lane line, lane line 2 is an overlapped lane line or a lane line called a repeat, and the target lane line combination may be, for example, each lane line in (b) in fig. 4, and the target lane line combination does not include lane line 1 and lane line 2, that is, the target lane line combination is a lane line combination configured by lane lines deleted from the lane line 1 and lane line 2 among the plurality of candidate lane lines.

In some embodiments, the second server may store the target lane line combination, and generate a map using the stored target lane line combination, for example, to generate a high-precision map. For example, when the second server acquires a high-accuracy map generation request, the second server acquires a stored combination of target lane lines, generates the latest high-accuracy map, and updates the existing high-accuracy map. The second server may push the latest high-precision map to the autonomous vehicle so that the autonomous vehicle moves in accordance with the latest high-precision map.

In some embodiments, the second server may store target lane line combinations corresponding to different time periods of the target traffic area, determine public lane lines of the target lane line combinations corresponding to the respective time periods, and generate the high-precision map using the respective public lane lines. The time periods corresponding to the target lane line combination are different, and the acquisition time of the image or point cloud from which the candidate lane line corresponding to the target lane line combination is derived is different. For example, the target lane line combination corresponding to the first day is obtained by processing a plurality of candidate lane lines obtained from the image acquired on the first day, and the target lane line combination corresponding to the second day is obtained by processing a plurality of candidate lane lines obtained from the image acquired on the second day. The common lane lines refer to lane lines included in the target lane line combinations respectively corresponding to the respective time periods.

The lane line detection processing method includes the steps of obtaining a plurality of candidate lane lines corresponding to a target traffic area, selecting at least two lane lines from the candidate lane lines to obtain an initial lane line combination, obtaining a spatial distribution relation among the lane lines in the initial lane line combination, obtaining lane line distribution scores corresponding to the spatial distribution relation, determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution scores, and optimizing the initial lane line combination based on the first combination detection score to obtain a target lane line combination. The first combination detection score is determined according to the lane line distribution score, and the lane line distribution score is obtained according to the spatial distribution relation among the lane lines, so that the first combination detection score can reflect the spatial distribution relation among the lane lines in the combination, the accuracy of the first combination detection score is improved, the initial lane line combination is optimized based on the first combination detection score, the target lane line combination is obtained, and the accuracy of the target lane line combination is improved.

The lane line detection processing method can be deployed after the lane line automatic extraction link so as to detect the automatically extracted lane lines. The lane line automatic extraction link refers to a process of extracting lane lines from the acquired images or point clouds. In order to improve the efficiency of lane line detection, the lane line detection processing method can be deployed in a cloud server, the cloud server can acquire lane lines obtained in lane line automatic extraction links of different batches, and the lane line detection processing method can be used for detecting the obtained lane lines in parallel. The lane line detection processing method can also be used in three-dimensional map production, and when a three-dimensional lane level road model is generated from a two-dimensional road level map in the three-dimensional map production, the lane line detection processing method can be used for detecting the two-dimensional lane line, so that incorrect lane lines are removed, and the accuracy of the three-dimensional map is improved.

In some embodiments, obtaining a spatial distribution relationship between lane lines in the initial lane line combination, and obtaining a lane line distribution score corresponding to the spatial distribution relationship includes: acquiring a spatial distribution relation between lane lines in the initial lane line combination; according to the distribution relation types, carrying out classification statistics on the spatial distribution relations to obtain the distribution relation quantity corresponding to each distribution relation type; and determining the distribution score of the lane line according to the distribution relation quantity corresponding to the distribution relation type.

The distribution relationship type may refer to a type of a spatial distribution relationship, that is, one spatial distribution relationship may correspond to one distribution relationship type, for example, "cross" corresponds to one distribution relationship type, and "overlap" corresponds to another distribution relationship type. The distribution relationship type may also be obtained by classifying the change direction of the combined detection score, for example, each spatial distribution relationship satisfying the lane line distribution relationship may be classified into a reward distribution relationship, and each spatial distribution relationship deviating from the lane line distribution relationship may be classified into a penalty distribution relationship, that is, the distribution relationship type may be any one of the reward distribution relationship and the penalty distribution relationship. The reward class distribution relationship may make the score better, and the penalty class distribution relationship may make the score worse. The bonus class distribution relationship is, for example, a "connection" relationship. The penalty class distribution relationship is, for example, a "cross" relationship or an "overlap" relationship.

The number of distribution relations corresponding to the distribution relation type refers to the number of lane line pairs corresponding to the distribution relation type, and the unit of the number is a pair. The number of the lane line pairs corresponding to the distribution relationship type may be the number of the lane line pairs corresponding to the same spatial distribution relationship in the initial lane line combination, for example, if there are 3 pairs of lane lines in the initial lane line combination and there is a "connection" relationship between 2 lane lines in each pair of lane lines, the number of the lane line pairs corresponding to the "connection" relationship is 3. The number of the lane line pairs corresponding to the distribution relationship type can also be the number of the lane line pairs corresponding to the spatial distribution relationship belonging to the reward distribution relationship in the initial lane line combination, or punishment type distribution relation, such as 3 pairs of lane lines in the initial lane line combination, 2 lane lines in the first pair of lane lines having a "connection" relation therebetween, 2 lane lines in the second pair of lane lines having an "overlap" relation therebetween, and 2 lane lines in the third pair of lane lines having a "cross" relation therebetween, since "overlap" and "cross" belong to the penalty class distribution relationship, "connect" belongs to the reward class distribution relationship, the number of the lane line pairs corresponding to the spatial distribution relationship of the reward type distribution relationship is 1, and the number of the lane line pairs corresponding to the spatial distribution relationship of the penalty type distribution relationship is 1+1, respectively.

Specifically, the server may perform classification statistics on the spatial distribution relations according to the distribution relation types, that is, count the number of spatial distribution relations belonging to the same classification relation type in each spatial distribution relation, and obtain the number of distribution relations corresponding to each distribution relation type. For example, when one distribution relationship type corresponds to one spatial distribution relationship, the second server may count the number of lane line pairs corresponding to the same spatial distribution relationship in the initial lane line combination as the distribution relationship number corresponding to the distribution relationship type, for example, the second server may set the number of lane line pairs corresponding to the "connection" relationship in the initial lane line combination as the distribution relationship number corresponding to the first distribution relationship type, set the number of lane line pairs corresponding to the "interaction" relationship as the distribution relationship number corresponding to the second distribution relationship type, and set the number of lane line pairs corresponding to the "overlap" relationship as the distribution relationship number corresponding to the third distribution relationship type. For example, in the initial lane combination, if the number of lane line pairs corresponding to the "cross" relationship is 3, the number of lane line pairs corresponding to the "overlap" relationship is 2, and the number of lane line pairs corresponding to the "connection" relationship is 5, the number of distribution relationships corresponding to each distribution relationship type is 3, 2, and 5, respectively.

In some embodiments, when one distribution relationship type corresponds to a plurality of spatial distribution relationships, for example, when the distribution relationship type includes a reward distribution relationship and a penalty distribution relationship, the second server may count, as the distribution relationship number corresponding to the distribution relationship type, the number of lane line pairs corresponding to the reward distribution relationship in the initial lane line combination and the number of lane line pairs corresponding to the penalty distribution relationship. For example, in the initial lane combination, if the number of lane line pairs corresponding to the "cross" relationship is 3, the number of lane line pairs corresponding to the "overlap" relationship is 2, and the number of lane line pairs corresponding to the "connection" relationship is 4, the number of distribution relationships corresponding to the penalty type distribution relationship type is 3+2, and the reward type distribution relationship type is 4.

In some embodiments, the second server may perform statistical calculation on the number of lane line pairs respectively corresponding to the distribution relationship types to obtain the lane line distribution score. The statistical calculation may include a summation calculation, such as a weighted summation calculation.

In this embodiment, the spatial distribution relationship between the lane lines in the initial lane line combination is obtained, the spatial distribution relationship is classified and counted according to the distribution relationship types, the distribution relationship number corresponding to each distribution relationship type is obtained, the lane line distribution score is determined according to the distribution relationship number corresponding to the distribution relationship type, and the accuracy of the lane line distribution score is improved.

In some embodiments, determining the lane line distribution score according to the number of distribution relations corresponding to the distribution relation types includes: acquiring the relation weight corresponding to the distribution relation type; and carrying out weighted summation according to the relation weight and the distribution relation quantity corresponding to the distribution relation type to obtain the lane line distribution score.

The relationship weight may be preset, or may be set as needed. When the distribution score of the lane line is the cost score, the corresponding relation weight of the distribution relation type of the reward class is a negative number, and the distribution relation type of the punishment class is a positive number. And when the lane line distribution score is the income score, the relationship weight corresponding to the distribution relationship type of the reward class is a positive number, and the relationship weight corresponding to the distribution relationship type of the penalty class is a negative number. The lane line distribution fraction and the relation weight form a positive correlation.

Specifically, the second server may perform weighted summation on the number of distribution relations corresponding to each distribution relation type according to the relation weight corresponding to each distribution relation type, so as to obtain the lane line distribution score. The lane line distribution score can be calculated by, for example, formula (2). Where, Uprior represents a lane line distribution score, Nconnected represents the number of lane line pairs connected to each other, Ncross represents the number of lane line pairs crossing each other, nparalel represents the number of lane line pairs overlapping each other, wc represents a relationship weight corresponding to a distribution relationship type of "connection", wr represents a relationship weight corresponding to a distribution relationship type of "crossing", and wp represents a relationship weight corresponding to a distribution relationship type of "overlapping".

Uprior＝wc*Nconnected+wr*Ncross+wp*Nparallel (2)

In this embodiment, the relationship weights corresponding to the distribution relationship types are obtained, and the weighting summation is performed according to the relationship weights and the distribution relationship numbers corresponding to the distribution relationship types to obtain the lane line distribution scores, so that the corresponding relationship weights can be set for different distribution relationship types, so that the relationship weights corresponding to the distribution relationship types satisfying the lane line distribution relationship are distinguished from the relationship weights corresponding to the distribution relationship types deviating from the lane line distribution relationship, the accuracy of the lane line distribution scores is improved, and the accuracy of lane line detection is improved.

In some embodiments, determining the lane line distribution score according to the number of distribution relations corresponding to the distribution relation types includes: when the spatial distribution relation corresponding to the distribution relation type meets the lane line distribution relation, obtaining a negative cost score according to the distribution relation quantity corresponding to the distribution relation type; the distribution relation quantity corresponding to the distribution relation type and the negative cost score form a negative correlation relation; and obtaining a lane line distribution score according to the negative cost score.

The lane line distribution relationship refers to a normal distribution relationship between lane lines, that is, a distribution relationship conforming to a lane line distribution rule. The requirement for the distribution relationship of the lane lines means that the spatial distribution relationship belongs to one of the distribution relationships of the lane lines, and the spatial distribution relationship meeting the distribution relationship of the lane lines can be set according to actual needs. For example, in general, if two lane lines are normal lane lines, the two lane lines are connected or parallel, and thus if the spatial distribution relationship is connected or parallel, the lane line distribution relationship is satisfied. The spatial distribution relationship corresponding to the distribution relationship type satisfies the lane line distribution relationship, which means that the spatial distribution relationship corresponding to the distribution relationship type belongs to the lane line distribution relationship. The cost scores include two categories, negative cost scores and positive cost scores. A negative cost score is a relative concept to a positive cost score. A negative cost score reduces the cost, i.e., is a negative number. A positive cost score increases the cost, i.e., is a positive number. The lane line distribution score may be obtained by adding the respective cost scores.

In this embodiment, when the distribution relationship corresponding to the distribution relationship type satisfies the distribution relationship of the lane lines, it indicates that the distribution relationship type is a good distribution relationship, and obtains the negative cost score according to the number of the distribution relationship corresponding to the distribution relationship type to obtain the distribution score of the lane lines.

In some embodiments, determining the lane line distribution score according to the number of distribution relations corresponding to the distribution relation types includes: when the spatial distribution relation corresponding to the distribution relation type deviates from the distribution relation of the lane lines, obtaining a forward cost score according to the quantity of the distribution relation corresponding to the distribution relation type; the distribution relation quantity corresponding to the distribution relation type and the forward cost score form a positive correlation; and obtaining a lane line distribution score according to the forward cost score.

The spatial distribution relationship corresponding to the distribution relationship type deviates from the distribution relationship of the lane lines, which means that the spatial distribution relationship corresponding to the distribution relationship type does not belong to the distribution relationship of the lane lines. For example, in general, if two lane lines are normal lane lines, the two lane lines are connected or parallel, not overlapped or crossed, and if the spatial distribution relationship is overlapped or crossed, the spatial distribution relationship is deviated from the lane line distribution relationship.

In some embodiments, the spatial distribution relationship between the lane lines in the initial lane line combination is a spatial distribution relationship that deviates from the lane line distribution relationship and a spatial distribution relationship that satisfies the lane line distribution relationship. The second server may obtain the number of distribution relationships corresponding to the type of distribution relationship in which the corresponding spatial distribution relationship satisfies the distribution relationship of the lane line, obtain a negative cost score, obtain the number of distribution relationships corresponding to the type of distribution relationship in which the corresponding spatial distribution relationship deviates from the distribution relationship of the lane line, obtain a positive cost score, and sum the negative cost score and the positive cost score to obtain the lane line distribution score.

In this embodiment, when the distribution relationship corresponding to the distribution relationship type deviates from the distribution relationship of the lane lines, it indicates that the distribution relationship type corresponds to a poor distribution relationship, and a forward cost score is obtained according to the number of the distribution relationship corresponding to the distribution relationship type, so as to obtain the distribution score of the lane lines.

In some embodiments, determining a first combination detection score corresponding to the initial lane line combination from the lane line distribution scores comprises: acquiring lane line attribute characteristics corresponding to each initial lane line in the initial lane line combination; acquiring a lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics; and determining a first combination detection score corresponding to the initial lane line combination according to the lane line attribute score and the lane line distribution score.

Wherein the initial lane line refers to a lane line in the initial lane line combination. The lane line attribute feature refers to a feature of a lane line, and may be plural, and may include, for example, a lane line length and a lane line angle. The lane line attribute score is determined based on one or more lane line attribute characteristics corresponding to the initialized lane line.

Specifically, the second server may obtain lane line attribute features corresponding to the initial lane line, determine attribute feature scores corresponding to the initial lane line according to the lane line attribute features, and determine lane line attribute scores corresponding to the initial lane line according to each attribute feature score. For example, the length fraction may be calculated according to the length of the lane line, the angle fraction may be calculated according to the lane line angle, and the lane line attribute fraction corresponding to the initial lane line may be determined according to the length fraction and the angle fraction. For example, one of the length score and the angle score may be selected as the lane line attribute score, or a weighted sum of the length score and the angle score may be calculated as the lane line attribute score.

In some embodiments, the second server may perform statistical calculation on the lane line attribute score to obtain an attribute score statistical value, and calculate a first combined detection score according to the attribute score statistical value and the lane line distribution score, for example, a result obtained by adding the attribute score statistical value and the lane line distribution score may be used as the first combined detection score.

In this embodiment, the lane line attribute features corresponding to each initial lane line in the initial lane line combination are obtained, the lane line attribute scores corresponding to the initial lane lines are obtained according to the lane line attribute features, and the first combination detection score corresponding to the initial lane line combination is determined according to the lane line attribute scores and the lane line distribution scores, so that the lane line attribute features and the spatial distribution relationship of the lane lines can be combined to determine the first combination detection score together, and the accuracy of the first combination detection score is improved.

In some embodiments, the lane line attribute characteristics include lane line length and lane line angle; the obtaining of the lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics, which is a cost score, includes: obtaining a length cost score corresponding to the initial lane line according to the length of the lane line; the length cost fraction and the length of the lane line form a negative correlation; obtaining an angle cost score corresponding to the initial lane line according to the lane line angle; the angle cost fraction and the lane line angle form a positive correlation; and selecting the score with the maximum price from the length cost score corresponding to the initial lane line and the angle cost score corresponding to the initial lane line as the attribute score of the lane line corresponding to the initial lane line.

The length cost fraction and the length of the lane line form a negative correlation relationship, the longer the length of the lane line is, the smaller the length cost fraction is, and the shorter the length of the lane line is, the larger the length cost fraction is. The length cost score may be any one of a negative cost score and a positive cost score, and the length cost score may be determined to be the negative cost score or the positive cost score according to a magnitude relationship between the lane line length and the lane line length threshold. The threshold value of the length of the lane line can be preset or can be set according to requirements. For example, the length of the lane line may be compared with a threshold of the length of the lane line, and when it is determined that the length of the lane line is greater than the threshold of the length of the lane line, the length cost score is determined as a negative cost score, and when it is determined that the length of the lane line is less than the threshold of the length of the lane line, the length cost score is determined as a positive cost score.

The angle cost score and the lane line angle form a positive correlation relationship, the larger the lane line angle is, the larger the angle cost score is, and the smaller the lane line angle is, the smaller the angle cost score is. The angle cost score may be any one of a negative cost score and a positive cost score, and the angle cost score may be determined to be the negative cost score or the positive cost score according to a magnitude relationship between the lane line angle and the lane line angle threshold. The lane line angle threshold may be preset or may be set as needed. For example, the lane line angle may be compared with a lane line angle threshold, when it is determined that the lane line angle is greater than the lane line angle threshold, the angle cost score is determined to be a positive cost score, and when it is determined that the lane line angle is less than the lane line angle threshold, the angle cost score is determined to be a negative cost score.

Specifically, the second server may compare the length cost score with the angle cost score, and when it is determined that the length cost score is greater than the angle cost score, take the length cost score as a lane line attribute score corresponding to the initial lane line. And when the length cost score is smaller than the angle cost score, taking the angle cost score as a lane line attribute score corresponding to the initial lane line. The lane line attribute score can be calculated, for example, by equation (3). udata ═ max (Ulength, Uangle) (3). Wherein, Ulength represents the length cost fraction, and urangle represents the angle cost fraction.

In some embodiments, when the length of the lane line is greater than the threshold length of the lane line, the second server may calculate a ratio of the threshold length of the lane line to the length of the lane line to obtain a first length ratio; calculating the difference between the first length ratio and the first value to obtain a first difference; and calculating a square value corresponding to the first difference, and taking the opposite number of the square value corresponding to the first difference as a length cost score. The first value may be preset or may be set as needed, and may be 1, for example. When the lane line length is greater than the lane line length threshold, the length cost score may be calculated, for example, using equation (4). Ulength ═ 1-t _ length/length ^2 (4). Where t _ length represents a lane line length threshold, length represents a lane line length, and 1 represents a first numerical value.

In some embodiments, when the lane line length is less than the lane line length threshold, the second server may calculate a ratio of the lane line length to the lane line length threshold to obtain a second length ratio; and calculating a square value corresponding to the second length ratio to obtain a ratio square result, and taking a result obtained by subtracting the ratio square result from the second numerical value as a length cost score. The second value may be preset or may be set according to needs, and the second value may be the same as the first value, and may be 1, for example. When the lane line length is less than the lane line length threshold, the length cost score may be calculated, for example, using equation (5). Wherein 1 represents a second numerical value. Ulength ═ 1- (length/t _ length) ^2 (5).

In some embodiments, when the lane line angle is greater than the lane line angle threshold, the second server may calculate a ratio of the lane line angle threshold to the lane line angle to obtain a first angle ratio; calculating the difference between the first angle ratio and the third value to obtain a second difference; and calculating a square value corresponding to the second difference, and taking the square value corresponding to the second difference as an angle cost score. The third value may be preset or may be set as needed, and may be 1, for example. When the lane angle is greater than the lane angle threshold, the angle cost score may be calculated, for example, using equation (6). Unangle ═ (1-t _ angle/std) ^2 (6). Where t _ angle represents the lane line angle threshold, std represents the lane line angle, and 1 represents a third value.

In some embodiments, when the lane line angle is smaller than the lane line angle threshold, the second server may calculate a ratio of the lane line angle to the lane line angle threshold to obtain a second angle ratio; and calculating a square value corresponding to the second angle ratio, and subtracting a fourth numerical value from the square value corresponding to the second angle ratio to obtain a result which is used as the angle cost score. The fourth value may be preset or may be set as needed, and may be 1, for example. When the lane angle is less than the lane angle threshold, the angle cost score may be calculated, for example, using equation (7). Wherein 1 represents a fourth numerical value. Unangle ═ (std/t _ angle) ^2-1 (7). The first, second, third and fourth values may be the same, e.g., all are 1.

In this embodiment, a length cost score corresponding to an initial lane line is obtained according to a length of the lane line, an angle cost score corresponding to the initial lane line is obtained according to an angle of the lane line, and a score with a maximum price is selected from the length cost score corresponding to the initial lane line and the angle cost score corresponding to the initial lane line as an attribute score of the lane line corresponding to the initial lane line, where the greater the length cost score is, the smaller the probability that the lane line satisfies the lane line distribution relationship is, the greater the angle cost score is, the smaller the probability that the lane line satisfies the lane line distribution relationship is, and for a lane line, when there is a mismatch between a length of the lane line and an angle of the lane line, the lane line is not qualified, so that the maximum score in the length cost score and the angle cost score is used as the attribute score of the lane line, and the characteristics of the lane line can be better reflected.

In some embodiments, optimizing the initial lane line combination based on the first combination detection score to obtain the target lane line combination includes: acquiring a current lane line from the initial lane line combination, and acquiring a second combination detection score obtained by subtracting the current lane line from the initial lane line combination; acquiring a score change value of the second combined detection score relative to the first combined detection score; and processing the current lane line in the initial lane line combination according to the score change value to obtain the target lane line combination.

The current lane line may be any one or more lane lines in the initial lane line combination, and the lane lines in the initial lane line combination may be sequentially used as the current lane line to determine whether to delete the lane line from the initial lane line combination. Or selecting the lane line meeting the conditions as the current lane line. For example, the second server may randomly obtain one or more lane lines from the initial lane line combination as the current lane line, or may sequence the lane lines in the initial lane line combination, and sequentially select lane lines from the initial lane line combination as the current lane line according to a sequencing result of the lane lines in the initial lane line combination. For example, when the attribute score of the lane line is the cost score, the second server may sort the lane lines in the initial lane line in the descending order of the attribute scores of the lane lines, and select the lane line sorted before the preset sort as the current lane line, so as to accelerate the convergence speed. The larger the lane line attribute score is, the farther forward the ranking is, the smaller the lane line attribute score is, and the farther backward the ranking is. The second server may select the lane lines in sequence from front to back according to the sequence of the lane lines as the current lane line. The earlier the ranking, the earlier the timing as the current lane line.

The second combination detection score is a combination detection score calculated from each lane line other than the current lane line in the initial lane line combination. The score change value refers to the result of subtracting the first combined detection score from the second combined detection score.

Specifically, the second server may use each lane line in the initial lane line combination as a current lane line, and obtain a second combination detection score corresponding to each current lane line. The second combined detection score corresponding to the current lane line refers to a combined detection score calculated according to each lane line except the current lane line in the initial lane line combination.

In some embodiments, the second server may determine an optimization processing operation corresponding to the current lane line according to the score change value, and process the current lane line in the initial lane line combination according to the optimization processing operation to obtain the target lane line combination. The optimization processing operation may be either one of reservation or deletion. Specifically, the second server may determine which lane lines in the initial lane line combination need to be reserved and which lane lines need to be deleted according to the score change value, delete the lane lines that need to be deleted from the initial lane line combination to obtain a processed lane line combination, and obtain the target lane line combination according to the processed lane line combination. For example, the second server may select one or more candidate lane directions different from lane lines in the processed lane line combination from the plurality of candidate lane lines, add the selected candidate lane lines to the processed lane line combination to serve as a next initial lane line combination, and perform optimization processing on the next initial lane line combination to obtain the target lane line combination.

In this embodiment, a current lane line is obtained from an initial lane line combination, a current lane line subtracted from the initial lane line combination is obtained, an obtained second combination detection score is obtained, a score change value of the second combination detection score relative to the first combination detection score is obtained, the current lane line in the initial lane line combination is processed according to the score change value, and a target lane line combination is obtained.

In some embodiments, processing the current lane line in the initial lane line combination according to the score change value to obtain the target lane line combination includes: processing the current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination; and selecting lane lines from the plurality of candidate lane lines, adding the lane lines into the processed lane line combination to obtain an initial lane line combination of the next round, returning to the steps of obtaining a spatial distribution relation among the lane lines in the initial lane line combination and obtaining lane line distribution scores corresponding to the spatial distribution relation until a stop condition is met, and obtaining a target lane line combination.

The processed lane line combination may be a lane line combination obtained by deleting a part of lane lines from the initial lane line combination according to the score change value. The stop condition refers to a condition for stopping the iteration. The stopping condition may be preset or may be set according to needs, for example, an iteration number threshold, which may be preset or determined according to the number of lane lines in the plurality of candidate lane lines, for example, the iteration number threshold may be in a positive correlation with the number of lane lines in the plurality of candidate lane lines, or the iteration number threshold may be set to a first number when the number of lane lines in the plurality of candidate lane lines is greater than the lane line number threshold, and the iteration number threshold may be set to a second number when the number of lane lines in the plurality of candidate lane lines is less than the lane line number threshold. The first number is greater than the second number. The first number and the second number may be preset, and the first number may be, for example, 50 times, and the second number may be, for example, 30 times.

Specifically, the second server may randomly extract any number or preset number of lane lines from the plurality of candidate lane lines, and add the extracted lane lines to the processed lane line combination to obtain the initial lane line combination of the next round. The preset number is a preset number, and may be preset as needed, for example. The second server may determine whether the current iteration number is equal to an iteration number threshold when obtaining the initial lane line combination of the next round, regard the processed lane line combination as a target lane line combination when the current iteration number is equal to the iteration number threshold, and return to the step of obtaining the spatial distribution relationship between the lane lines in the initial lane line combination and obtaining the lane line distribution score corresponding to the spatial distribution relationship when the current iteration number is less than the iteration number threshold.

In some embodiments, in order to avoid lane line duplication, the second server may select a lane line different from the lane lines included in the processed lane line combination from the plurality of candidate lane lines, that is, select a lane line not belonging to the processed lane line combination from the plurality of candidate lane lines, and add the lane line to the processed lane line combination to obtain the initial lane line combination of the next round.

In some embodiments, the second server may repeat multiple rounds of the step of deriving the initial lane line combination for the next round from the initial lane line combination. The number of lane lines added to the processed lane line combination may be decreasing each round, for example, may be exponentially decreasing.

In some embodiments, since the first combined detection score may be represented as a result of adding the attribute score statistic and the lane line score, the lane lines included in the initial lane line combination may be adjusted such that the combined detection score calculated by the initial lane line combination is minimized, and the initial lane line combination score corresponding to the minimized combined detection score is taken as the processed lane line combination. That is, a combination obtained by the lane lines in the initial lane line combination satisfying the formula (8) may be used as the processed lane line combination.

minU＝min{∑max(Ulength,Uangle)+wc*Nconnected+wr*Ncross+wp*Nparallel} (8)

In this embodiment, a current lane line in the initial lane line combination is processed according to the score change value to obtain a processed lane line combination, a lane line is selected from the multiple candidate lane lines and added to the processed lane line combination to obtain an initial lane line combination of a next round, the steps of obtaining a spatial distribution relationship between lane lines in the initial lane line combination and obtaining lane line distribution scores corresponding to the spatial distribution relationship are returned until a stop condition is met to obtain a target lane line combination, and lane lines deviating from the lane line distribution relationship among the multiple candidate lane lines can be filtered out through multiple iterations, so that the obtained target lane line combination includes lane lines meeting the lane line distribution relationship, and the accuracy of the target lane line combination is improved.

In some embodiments, the score change value is a cost change value, and the processing a current lane line in the initial lane line combination according to the score change value, to obtain a processed lane line combination includes: when the price change value is larger than the first price threshold value, acquiring the corresponding acceptance probability of the current lane line; determining the optimization processing operation corresponding to the current lane line from the initial lane line combination according to the receiving probability; and processing the current lane line in the initial lane line combination according to the optimization processing operation to obtain the processed lane line combination.

The first price threshold may be preset, or may be set as needed, and may be 0, for example.

The acceptance probability refers to the probability that the lane line is retained in the combination. The acceptance probability may be preset or may be set according to needs, the acceptance probability in the same iteration is fixed, and as the number of iterations increases, the acceptance probability continuously decreases, for example, decreases exponentially, so that as the lane line combination becomes more and more accurate, the probability that the lane line is retained is smaller. The acceptance probability is smaller than the preset probability, the preset probability may be set according to needs, for example, may be 0.5, and the acceptance probability may be 0.3, for example.

Specifically, when the cost change value is larger than the first cost threshold, indicating that the current lane line is deleted from the initial lane line combination, the combination detection score of the initial lane line combination may be changed toward a larger direction, that is, the distribution of the lane lines in the initial lane line combination may be changed toward a worse direction.

In some embodiments, the second server may generate a random number between 0 and 1, compare the generated random number with the acceptance probability, determine that the optimization processing operation corresponding to the current lane line is deletion when the generated random number is greater than the acceptance probability, delete the lane line from the initial lane line combination, determine that the optimization processing operation corresponding to the current lane line is retention when the generated random number is less than the acceptance probability, and retain the lane line in the initial lane line combination to obtain the processed lane line combination.

In some embodiments, the acceptance probability and the cost change value are in a negative correlation relationship, and the larger the cost change value is, the smaller the acceptance probability is, and the smaller the cost change value is, the larger the acceptance probability is.

In this embodiment, when the cost variation value is greater than the first cost threshold, the current lane line may also satisfy the lane line distribution relationship, so that the optimization processing operation corresponding to the current lane line is determined from the initial lane line combination according to the acceptance probability, and the current lane line in the initial lane line combination is processed according to the optimization processing operation, so that the lane line satisfying the lane line distribution relationship can be prevented from being deleted, and the accuracy of lane line detection is improved.

In some embodiments, the score change value is a cost change value, and the processing a current lane line in the initial lane line combination according to the score change value, to obtain a processed lane line combination includes: and when the cost change value is smaller than the first cost threshold value, deleting the current lane line from the initial lane line combination to obtain a processed lane line combination.

Specifically, when the value of change of the representative is smaller than the first value threshold, which explains a case where the current lane line is deleted from the initial lane line combination, it is possible to make the combination detection score of the initial lane line combination change toward a smaller direction, i.e., to make the distribution of the lane lines in the initial lane line combination change toward a better direction. Therefore, when the cost variation value is smaller than the first cost threshold, the second server may delete the current lane line having the cost variation value smaller than the first cost threshold from the initial lane line combination, to obtain the processed lane line combination.

In this embodiment, when the value of change is smaller than the first value threshold, the current lane line is deleted from the initial lane line combination to obtain a processed lane line combination, so that the initial lane line combination changes toward a direction in which the lane line distribution becomes good.

In some embodiments, as shown in fig. 8, a lane line detection processing method is provided, which includes the following steps:

s802, a plurality of candidate lane lines corresponding to the target traffic area are obtained.

S804, at least two lane lines are selected from the candidate lane lines to obtain an initial lane line combination.

S806, obtaining the spatial distribution relation among the lane lines in the initial lane line combination, and carrying out classification statistics on the spatial distribution relation according to the distribution relation types to obtain the distribution relation quantity corresponding to each distribution relation type.

And S808, determining the distribution score of the lane line according to the distribution relation number corresponding to the distribution relation type.

The second server may obtain the relationship weight corresponding to the distribution relationship type, and perform weighted summation according to the relationship weight and the distribution relationship number corresponding to the distribution relationship type to obtain the lane line distribution score.

The second server can also obtain a negative cost score according to the number of the distribution relations corresponding to the distribution relation types when the spatial distribution relations corresponding to the distribution relation types meet the distribution relations of the lane lines; the distribution relation quantity corresponding to the distribution relation type and the negative cost score form a negative correlation relation; and obtaining a lane line distribution score according to the negative cost score. When the spatial distribution relation corresponding to the distribution relation type deviates from the distribution relation of the lane lines, obtaining a forward cost score according to the quantity of the distribution relation corresponding to the distribution relation type; the distribution relation quantity corresponding to the distribution relation type and the forward cost score form a positive correlation; and obtaining a lane line distribution score according to the forward cost score.

S810, obtaining lane line attribute characteristics corresponding to each initial lane line in the initial lane line combination; and acquiring a lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics.

The second server can obtain the length cost score corresponding to the initial lane line according to the length of the lane line; the length cost fraction and the length of the lane line form a negative correlation; obtaining an angle cost score corresponding to the initial lane line according to the lane line angle; the angle cost fraction and the lane line angle form a positive correlation; and selecting the score with the maximum price from the length cost score corresponding to the initial lane line and the angle cost score corresponding to the initial lane line as the attribute score of the lane line corresponding to the initial lane line.

S812, determining a first combination detection score corresponding to the initial lane line combination according to the lane line attribute score and the lane line distribution score.

S814, obtaining a current lane line from the initial lane line combination, and obtaining a second combination detection score obtained by subtracting the current lane line from the initial lane line combination; a score change value of the second combined detection score relative to the first combined detection score is obtained.

And S816, processing the current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination.

When the price change value is larger than the first price threshold value, acquiring the corresponding acceptance probability of the current lane line; determining the optimization processing operation corresponding to the current lane line from the initial lane line combination according to the receiving probability; and processing the current lane line in the initial lane line combination according to the optimization processing operation to obtain the processed lane line combination. And when the cost change value is smaller than the first cost threshold value, deleting the current lane line from the initial lane line combination to obtain a processed lane line combination.

S818 determines whether the stop condition is satisfied.

In which, for example, it is determined whether the number of iterations reaches a specified number of iterations.

And S820, if not, selecting a lane line from the plurality of candidate lane lines, and adding the lane line into the processed lane line combination to obtain the initial lane line combination of the next round. And returning to the step of obtaining the spatial distribution relation among the lane lines in the initial lane line combination and obtaining the lane line distribution fraction corresponding to the spatial distribution relation.

And S822, if so, obtaining the target lane line combination. Namely, the processed lane line combination is used as the target lane line combination.

The lane line detection processing method provided by the application is exemplified as follows: assume that the number of lane lines in the plurality of lane line candidates is M. And randomly selecting M lane lines from the M candidate lane lines to obtain an initial lane line combination and an initial remaining lane line combination, wherein the initial remaining lane line combination is formed by lane lines except the initial lane line combination in the candidate lane lines, and the remaining lane line combination comprises M-M lane lines. M is less than M.

In the 0 th iteration, a cost score U0 corresponding to the initial lane-line combination is calculated, and optimization processing is performed on the initial lane-line combination, for example, n0 lane lines are deleted from the initial lane-line combination as a result of the optimization processing, so that a processed lane-line combination is obtained. And selecting P0 lane lines from the n0 deleted lane lines and the initial remaining lane line combination, and adding the P0 lane lines into the processed lane line combination to obtain an initial lane line combination corresponding to the 1 st iteration. And adding the deleted n0 lane lines into the initial remaining lane line combination, and deleting the selected P0 lane lines from the initial remaining lane line combination to obtain the remaining lane line combination corresponding to the 0 th iteration.

In the 1 st iteration, the cost score U1 of the initial lane line combination corresponding to the 1 st iteration is calculated, and the initial lane line combination corresponding to the 1 st iteration is optimized, for example, n1 lane lines are deleted from the initial lane line combination corresponding to the 1 st iteration as a result of the optimization, so as to obtain the processed lane line combination. And selecting P1 lane lines from the n0 deleted lane lines and the remaining lane line combination corresponding to the 0 th iteration, and adding the P1 lane lines into the processed lane line combination to obtain an initial lane line combination corresponding to the 2 nd iteration. Adding the deleted n1 lane lines into the remaining lane line combination corresponding to the 0 th iteration, and deleting the selected P1 lane lines from the remaining lane line combination corresponding to the 0 th iteration to obtain the remaining lane line combination corresponding to the 1 st iteration. Wherein, P1 is different from P0, and P1 is the result of P0 decreasing according to the index. As the number of iterations increases, the number of lane lines per increment, for example, the 0 th increment P0, decreases exponentially. The number of subsequent iterations is the same as the method for the 0 th iteration and the 1 st iteration.

And when the iteration times reach the specified iteration times, taking the processed lane line combination obtained by the last iteration as a target lane line combination to obtain the required lane line. The process of cost function reduction in the iteration process, the change of the number of lane lines, and the change of the number of lane lines increased and decreased in each iteration process are respectively shown in fig. 5, fig. 6, and fig. 7. The lane line number variation means that the number in the processed lane line combination varies with the number of iterations.

The lane line detection processing method can be understood as a data regularization method based on geometric constraint, a lane line network model based on a linear set object is established, and the geometric and topological relation of lane lines is simulated; the data abnormity can be abstracted as the conflict between object-level geometries, such as deformation, intersection and the like, and the data abnormity can be detected in a self-adaptive manner by adopting a cost function method, so that irregular lane lines in automatic identification can be obviously removed in a self-adaptive manner; the model and data matching problem and the data abnormity detection are decoupled, namely a method for detecting the lane line is decoupled from the lane line, and the lane line detection processing method does not depend on a specific lane line; the lane line detection result obtained through deep learning detection can be automatically corrected, manual participation is reduced, and the production efficiency and the running stability of the production platform are improved. The lane lines which do not meet the geometric constraint in the lane lines can be detected in a self-adaptive mode, the geometric constraint can be expanded along with maintenance, and the general data processing method for mass production application is achieved.

It should be understood that although the various steps in the flow charts of fig. 2-8 are shown in order as indicated by the arrows, the steps are not necessarily performed in order as indicated by the arrows. The steps are not performed in the exact order shown and described, and may be performed in other orders, unless explicitly stated otherwise. Moreover, at least some of the steps in fig. 2-8 may include multiple steps or multiple stages, which are not necessarily performed at the same time, but may be performed at different times, which are not necessarily performed in sequence, but may be performed in turn or alternately with other steps or at least some of the other steps.

In some embodiments, as shown in fig. 9, there is provided a lane line detection processing apparatus, which may be a part of a computer device using a software module or a hardware module, or a combination of the two, and specifically includes: a lane line candidate obtaining module 902, an initial lane line combination obtaining module 904, a lane line distribution score obtaining module 906, a first combination detection score determining module 908, and a target lane line combination obtaining module 910, wherein:

a lane line candidate obtaining module 902, configured to obtain a plurality of lane line candidates corresponding to the target traffic area.

An initial lane line combination obtaining module 904, configured to select at least two lane lines from the multiple candidate lane lines to obtain an initial lane line combination.

A lane line distribution score obtaining module 906, configured to obtain a spatial distribution relationship between lane lines in the initial lane line combination, and obtain a lane line distribution score corresponding to the spatial distribution relationship.

The first combination detection score determining module 908 is configured to determine a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score.

And a target lane combination obtaining module 910, configured to perform optimization processing on the initial lane combination based on the first combination detection score to obtain a target lane combination.

In some embodiments, the lane line distribution score acquisition module 906 includes: a spatial distribution relation acquisition unit for acquiring a spatial distribution relation between lane lines in the initial lane line combination; a distribution relation quantity obtaining unit, configured to perform classification statistics on the spatial distribution relations according to the distribution relation types to obtain distribution relation quantities corresponding to the distribution relation types; and the lane line distribution score determining unit is used for determining the lane line distribution score according to the distribution relation quantity corresponding to the distribution relation type.

In some embodiments, the lane line distribution score obtaining unit is further configured to obtain a negative cost score according to the number of distribution relations corresponding to the distribution relation types when the spatial distribution relation corresponding to the distribution relation type satisfies the lane line distribution relation; the distribution relation quantity corresponding to the distribution relation type and the negative cost score form a negative correlation relation; and obtaining a lane line distribution score according to the negative cost score.

In some embodiments, the lane line distribution score obtaining unit is further configured to obtain a forward cost score according to the number of distribution relations corresponding to the type of distribution relation when the spatial distribution relation corresponding to the type of distribution relation deviates from the lane line distribution relation; the distribution relation quantity corresponding to the distribution relation type and the forward cost score form a positive correlation; and obtaining a lane line distribution score according to the forward cost score.

In some embodiments, the first combined detection score determination module 908 comprises:

and the lane attribute characteristic acquisition unit is used for acquiring lane attribute characteristics corresponding to each initial lane line in the initial lane line combination.

And the lane line attribute score acquiring unit is used for acquiring the lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics.

And the first combination detection score obtaining unit is used for determining a first combination detection score corresponding to the initial lane line combination according to the lane line attribute score and the lane line distribution score.

In some embodiments, the lane line attribute characteristics include lane line length and lane line angle; the lane line attribute score is a cost score, and the lane line attribute score acquisition unit is also used for acquiring a length cost score corresponding to the initial lane line according to the length of the lane line; the length cost fraction and the length of the lane line form a negative correlation; obtaining an angle cost score corresponding to the initial lane line according to the lane line angle; the angle cost fraction and the lane line angle form a positive correlation; and selecting the score with the maximum price from the length cost score corresponding to the initial lane line and the angle cost score corresponding to the initial lane line as the attribute score of the lane line corresponding to the initial lane line.

In some embodiments, the target lane line combination derivation module 910 includes:

and the second combination detection score acquisition unit is used for acquiring the current lane line from the initial lane line combination, and acquiring a second combination detection score obtained by subtracting the current lane line from the initial lane line combination.

A score change value acquisition unit for acquiring a score change value of the second combination detection score with respect to the first combination detection score.

And the target lane combination obtaining unit is used for processing the current lane in the initial lane combination according to the fractional change value to obtain the target lane combination.

In some embodiments, the target lane line combination obtaining unit is further configured to process a current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination; selecting a lane line from the plurality of candidate lane lines, and adding the lane line into the processed lane line combination to obtain an initial lane line combination of the next round; and returning to the steps of obtaining the spatial distribution relation among the lane lines in the initial lane line combination and obtaining the lane line distribution fraction corresponding to the spatial distribution relation until the stop condition is met, and obtaining the target lane line combination.

In some embodiments, the score variation value is a cost variation value, and the target lane line combination obtaining unit is further configured to obtain an acceptance probability corresponding to the current lane line when the cost variation value is greater than a first cost threshold; determining the optimization processing operation corresponding to the current lane line from the initial lane line combination according to the receiving probability; and processing the current lane line in the initial lane line combination according to the optimization processing operation to obtain the processed lane line combination.

In some embodiments, the score variation value is a cost variation value, and the target lane line combination obtaining unit is further configured to delete the current lane line from the initial lane line combination when the cost variation value is smaller than a first cost threshold value, so as to obtain a processed lane line combination.

For the specific definition of the lane line detection processing device, the above definition of the lane line detection processing method may be referred to, and is not described herein again. Each module in the lane line detection processing device may be wholly or partially implemented by software, hardware, or a combination thereof. The modules can be embedded in a hardware form or independent from a processor in the computer device, and can also be stored in a memory in the computer device in a software form, so that the processor can call and execute operations corresponding to the modules.

In some embodiments, a computer device is provided, which may be a server, the internal structure of which may be as shown in fig. 10. The computer device includes a processor, a memory, and a network interface connected by a system bus. Wherein the processor of the computer device is configured to provide computing and control capabilities. The memory of the computer device comprises a nonvolatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of an operating system and computer programs in the non-volatile storage medium. The database of the computer device is used for storing data involved in the lane line detection processing method. The network interface of the computer device is used for communicating with an external terminal through a network connection. The computer program is executed by a processor to implement a lane line detection processing method.

Those skilled in the art will appreciate that the architecture shown in fig. 10 is merely a block diagram of some of the structures associated with the disclosed aspects and is not intended to limit the computing devices to which the disclosed aspects apply, as particular computing devices may include more or less components than those shown, or may combine certain components, or have a different arrangement of components.

In some embodiments, there is further provided a computer device comprising a memory and a processor, the memory having stored therein a computer program, the processor implementing the steps of the above method embodiments when executing the computer program.

In some embodiments, a computer-readable storage medium is provided, in which a computer program is stored, which computer program, when being executed by a processor, carries out the steps of the above-mentioned method embodiments.

In some embodiments, a computer program product or computer program is provided that includes computer instructions stored in a computer-readable storage medium. The computer instructions are read by a processor of a computer device from a computer-readable storage medium, and the computer instructions are executed by the processor to cause the computer device to perform the steps in the above-mentioned method embodiments.

It will be understood by those skilled in the art that all or part of the processes of the methods of the embodiments described above can be implemented by hardware instructions of a computer program, which can be stored in a non-volatile computer-readable storage medium, and when executed, can include the processes of the embodiments of the methods described above. Any reference to memory, storage, database or other medium used in the embodiments provided herein can include at least one of non-volatile and volatile memory. Non-volatile Memory may include Read-Only Memory (ROM), magnetic tape, floppy disk, flash Memory, optical storage, or the like. Volatile Memory can include Random Access Memory (RAM) or external cache Memory. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM), among others.

The technical features of the above embodiments can be arbitrarily combined, and for the sake of brevity, all possible combinations of the technical features in the above embodiments are not described, but should be considered as the scope of the present specification as long as there is no contradiction between the combinations of the technical features.

The above-mentioned embodiments only express several embodiments of the present application, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the concept of the present application, which falls within the scope of protection of the present application. Therefore, the protection scope of the present patent shall be subject to the appended claims.

Claims

1. A lane line detection processing method is characterized by comprising the following steps:

obtaining a plurality of candidate lane lines corresponding to a target traffic area;

selecting at least two lane lines from the candidate lane lines to obtain an initial lane line combination;

acquiring a spatial distribution relationship among the lane lines in the initial lane line combination, wherein the spatial distribution relationship comprises at least one of a cross relationship, an overlapping relationship or a connection relationship, the cross relationship refers to that the tail part of the lane line is overlapped with the head part of the lane line, the overlapping relationship refers to that a part of or all the regions of two or more lane lines are overlapped, and the connection relationship refers to that the two or more lane lines are mutually crossed;

according to the distribution relation types, carrying out classification statistics on the spatial distribution relations to obtain the distribution relation quantity corresponding to each distribution relation type;

determining a lane line distribution score according to the distribution relation quantity corresponding to the distribution relation type;

determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score;

and optimizing the initial lane line combination based on the first combination detection score to obtain a target lane line combination.

2. The method of claim 1, wherein determining a first combination detection score corresponding to the initial lane line combination from the lane line distribution scores comprises:

and carrying out statistical calculation on the distribution scores of the lane lines, and taking the calculated result as a first combination detection score corresponding to the initial lane line combination.

3. The method according to claim 1, wherein the determining the lane line distribution score according to the distribution relationship number corresponding to the distribution relationship type comprises:

acquiring the relation weight corresponding to the distribution relation type;

and carrying out weighted summation according to the relation weight and the distribution relation quantity corresponding to the distribution relation type to obtain the lane line distribution score.

4. The method according to claim 1, wherein the determining the lane line distribution score according to the distribution relationship number corresponding to the distribution relationship type comprises:

when the spatial distribution relation corresponding to the distribution relation type meets the distribution relation of the lane lines, obtaining a negative cost score according to the quantity of the distribution relation corresponding to the distribution relation type; the distribution relation quantity corresponding to the distribution relation type and the negative cost score form a negative correlation relation;

and obtaining the lane line distribution score according to the negative cost score.

5. The method according to claim 1, wherein the determining the lane line distribution score according to the distribution relationship number corresponding to the distribution relationship type comprises:

when the spatial distribution relation corresponding to the distribution relation type deviates from the distribution relation of the lane lines, obtaining a forward cost score according to the quantity of the distribution relation corresponding to the distribution relation type; wherein, the distribution relation quantity corresponding to the distribution relation type and the forward cost score form a positive correlation;

and obtaining the lane line distribution score according to the forward cost score.

6. The method of claim 1, wherein determining a first combination detection score corresponding to the initial lane line combination from the lane line distribution scores comprises:

acquiring the attribute characteristics of the lane lines corresponding to each initial lane line in the initial lane line combination;

acquiring a lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics;

and determining a first combination detection score corresponding to the initial lane line combination according to the lane line attribute score and the lane line distribution score.

7. The method of claim 6, wherein the lane line attribute features include lane line length and lane line angle; the obtaining of the lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics includes:

obtaining a length cost score corresponding to the initial lane line according to the length of the lane line; the length cost fraction is in a negative correlation with the lane line length;

obtaining an angle cost score corresponding to the initial lane line according to the lane line angle; the angle cost fraction and the lane line angle form a positive correlation;

and selecting the score with the maximum substitution value from the length cost score corresponding to the initial lane line and the angle cost score corresponding to the initial lane line as the attribute score of the lane line corresponding to the initial lane line.

8. The method of claim 1, wherein the optimizing the initial lane line combination based on the first combination detection score to obtain a target lane line combination comprises:

acquiring a current lane line from the initial lane line combination, and acquiring a second combination detection score obtained by subtracting the current lane line from the initial lane line combination;

obtaining a score change value of the second combined detection score relative to the first combined detection score;

and processing the current lane line in the initial lane line combination according to the score change value to obtain a target lane line combination.

9. The method of claim 8, wherein the processing a current lane line of the initial lane line combination according to the fractional change value to obtain a target lane line combination comprises:

processing the current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination;

and selecting lane lines from the plurality of candidate lane lines, adding the lane lines into the processed lane line combination to obtain an initial lane line combination of the next round, returning to the steps of obtaining a spatial distribution relation among the lane lines in the initial lane line combination and obtaining lane line distribution scores corresponding to the spatial distribution relation until a stop condition is met, and obtaining a target lane line combination.

10. The method of claim 9, wherein the score change value is a cost change value, and the processing a current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination comprises:

when the cost change value is larger than a first cost threshold value, acquiring the corresponding acceptance probability of the current lane line;

determining the optimization processing operation corresponding to the current lane line from the initial lane line combination according to the receiving probability;

and processing the current lane line in the initial lane line combination according to the optimization processing operation to obtain a processed lane line combination.

11. The method of claim 9, wherein the score change value is a cost change value, and the processing a current lane line in the initial lane line combination according to the score change value to obtain a processed lane line combination comprises:

and when the cost change value is smaller than a first cost threshold value, deleting the current lane line from the initial lane line combination to obtain a processed lane line combination.

12. A lane line detection processing apparatus, characterized in that the apparatus comprises:

the candidate lane line acquisition module is used for acquiring a plurality of candidate lane lines corresponding to the target traffic area;

an initial lane line combination obtaining module, configured to select at least two lane lines from the multiple candidate lane lines to obtain an initial lane line combination;

the lane line distribution score acquisition module is used for acquiring a spatial distribution relationship among lane lines in the initial lane line combination, wherein the spatial distribution relationship comprises at least one of a cross relationship, an overlapping relationship or a connection relationship, the cross relationship refers to that the tail part of a lane line is overlapped with the head part of the lane line, the overlapping relationship refers to that partial areas or all areas of two or more lane lines are overlapped, and the connection relationship refers to that two or more lane lines are mutually crossed; according to the distribution relation types, carrying out classification statistics on the spatial distribution relations to obtain the distribution relation quantity corresponding to each distribution relation type; determining a lane line distribution score according to the distribution relation quantity corresponding to the distribution relation type;

the first combination detection score determining module is used for determining a first combination detection score corresponding to the initial lane line combination according to the lane line distribution score;

and the target lane combination obtaining module is used for optimizing the initial lane combination based on the first combination detection score to obtain a target lane combination.

13. The apparatus of claim 12, wherein the first combined detection score determination module is further configured to:

14. The apparatus of claim 12, wherein the lane line distribution score obtaining module is further configured to:

acquiring the relation weight corresponding to the distribution relation type;

15. The apparatus of claim 12, wherein the lane line distribution score obtaining module is further configured to:

16. The apparatus of claim 12, wherein the lane line distribution score obtaining module is further configured to:

17. The apparatus of claim 12, wherein the first combined detection score determining module comprises:

a lane attribute feature obtaining unit, configured to obtain lane attribute features corresponding to each initial lane in the initial lane combination;

the lane line attribute score acquiring unit is used for acquiring a lane line attribute score corresponding to the initial lane line according to the lane line attribute characteristics;

18. The apparatus of claim 17, wherein the lane line attribute characteristics include lane line length and lane line angle; the lane line attribute score is a cost score, and the lane line attribute score obtaining unit is further configured to:

19. The apparatus of claim 12, wherein the target lane line combination derivation module comprises:

a second combination detection score obtaining unit, configured to obtain a current lane line from the initial lane line combination, and obtain a second combination detection score obtained by subtracting the current lane line from the initial lane line combination;

a score change value acquisition unit configured to acquire a score change value of the second combination detection score with respect to the first combination detection score;

and the target lane line combination obtaining unit is used for processing the current lane line in the initial lane line combination according to the fractional change value to obtain the target lane line combination.

20. The apparatus of claim 19, wherein the target lane line combination deriving unit is further configured to:

21. The apparatus of claim 20, wherein the score change value is a cost change value, and wherein the target lane line combination deriving unit is further configured to:

22. The apparatus of claim 20, wherein the score change value is a cost change value, and wherein the target lane line combination deriving unit is further configured to:

23. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor realizes the steps of the method of any one of claims 1 to 11 when executing the computer program.

24. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 11.

25. A computer program product comprising a computer program, characterized in that the computer program realizes the steps of the method of any one of claims 1 to 11 when executed by a processor.