WO2021208110A1 - 车道线识别异常事件确定方法、车道线识别装置及系统 - Google Patents

车道线识别异常事件确定方法、车道线识别装置及系统 Download PDF

Info

Publication number
WO2021208110A1
WO2021208110A1 PCT/CN2020/085470 CN2020085470W WO2021208110A1 WO 2021208110 A1 WO2021208110 A1 WO 2021208110A1 CN 2020085470 W CN2020085470 W CN 2020085470W WO 2021208110 A1 WO2021208110 A1 WO 2021208110A1
Authority
WO
WIPO (PCT)
Prior art keywords
lane line
line recognition
time stamp
confidence level
vehicle
Prior art date
Application number
PCT/CN2020/085470
Other languages
English (en)
French (fr)
Inventor
谢瑜
俞佳伟
佘晓丽
Original Assignee
华为技术有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 华为技术有限公司 filed Critical 华为技术有限公司
Priority to PCT/CN2020/085470 priority Critical patent/WO2021208110A1/zh
Priority to EP20931648.8A priority patent/EP4130668A4/en
Priority to CN202080004432.3A priority patent/CN112639765B/zh
Publication of WO2021208110A1 publication Critical patent/WO2021208110A1/zh
Priority to US17/967,376 priority patent/US20230034574A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/98—Detection or correction of errors, e.g. by rescanning the pattern or by human intervention; Evaluation of the quality of the acquired patterns
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • G06V20/588—Recognition of the road, e.g. of lane markings; Recognition of the vehicle driving pattern in relation to the road
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2465—Query processing support for facilitating data mining operations in structured databases
    • G—PHYSICS
    • G01—MEASURING; TESTING
    • G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
    • G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
    • G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/23—Updating
    • G06F16/2365—Ensuring data consistency and integrity
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/24—Querying
    • G06F16/245—Query processing
    • G06F16/2458—Special types of queries, e.g. statistical queries, fuzzy queries or distributed queries
    • G06F16/2474—Sequence data queries, e.g. querying versioned data
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
    • G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
    • G06F16/29—Geographical information databases
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/762—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using clustering, e.g. of similar faces in social networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V10/00—Arrangements for image or video recognition or understanding
    • G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
    • G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N20/00—Machine learning
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the present application relates to the field of automatic driving, and in particular, to a method for determining an abnormal event of lane line recognition, a device and system for lane line recognition.
  • the core technologies in the field of automatic driving include intelligent environment perception, automatic navigation and positioning, driving behavior decision-making, and intelligent path planning and control.
  • the vehicle needs to detect the lane line in the road, and then the vehicle completes the driving behavior based on the detected lane line and the opened function.
  • the lane line recognition result is generally obtained by the lane line recognition device installed on the vehicle side based on the algorithm to process the image information obtained by the vehicle during the driving process.
  • the algorithm often adopts the deep learning scheme based on neural network (such as but not limited to CNN). (Convolutional Neural Network) etc.).
  • lidar method used to identify lane lines, but if the lidar method is used, it is limited by the reflectivity of the lane line. If the reflectivity of the lane line in the laser point cloud is low, it will affect the generation of the lane line Accuracy: If the data volume of the laser point cloud is increased, a large amount of storage space is required and a large amount of computing time is consumed, which makes the application of lidar methods on the car side expensive.
  • various embodiments of the present application provide a method for determining the confidence level of lane line recognition, a lane line recognition device, a lane line recognition system, and a computer-readable storage medium.
  • an embodiment of the present application provides a method for determining the confidence level of lane line recognition.
  • the method may specifically include: first determining the updated lane line recognition confidence level based on the posterior lane line recognition result; and then based on the updated lane line recognition The confidence level constructs the true value of the lane line; finally, the lane line recognition abnormal event is determined based on the lane line recognition result and the true value of the lane line, and the posterior lane line recognition result includes the lane line recognition confidence obtained after the vehicle is driven. Spend.
  • This application updates the previously obtained lane line recognition confidence level based on the posterior lane line recognition result, which improves the accuracy and reliability of the lane line recognition confidence level; and uses the updated lane line recognition confidence level to construct the lane line truth
  • Lane line recognition abnormal events can be obtained through the constructed lane line true value and lane line recognition confidence. These abnormal events can be used to further train the lane line recognition algorithm, thereby improving the recognition accuracy and precision of the lane line recognition algorithm.
  • the posterior lane line recognition result and the posterior inertial navigation data are aligned based on the timestamp; the posterior inertial navigation data includes the inertial navigation obtained after the vehicle is driven. data.
  • the spatial position of the vehicle and the lane line recognition results at different moments can be determined based on the time axis.
  • Lane line recognition results include: vehicle position, lane line recognition confidence and lane line recognition length.
  • the updated information corresponding to the timestamp is determined based on at least two lane line recognition confidences in the posterior lane line recognition result corresponding to the timestamp.
  • Lane line recognition confidence level may include: for any time stamp, determining a time stamp set corresponding to the any time stamp, the time stamp set including one or more time stamps, and lane lines of the one or more time stamps.
  • the recognition length includes the vehicle position corresponding to any one of the time stamps, and the lane line recognition confidence level corresponding to each time stamp in the time stamp set is obtained to form a lane line recognition confidence set, and the lane line recognition Obtain at least two from the confidence set and sum them to obtain the updated lane line recognition confidence corresponding to any one of the timestamps.
  • the technical step may include: summing all lane line recognition confidences in the lane line recognition confidence set to obtain an updated lane line recognition confidence corresponding to any of the timestamps Spend.
  • the updated lane line recognition confidence level of the time stamp is determined by summing the lane line recognition confidence level of at least two/more lane line recognition ranges including its time stamp, so that The error of the lane line recognition confidence level caused by the recognition fluctuation of a single time stamp is suppressed, so for any time stamp, the accuracy/reliability of the updated lane line recognition confidence level is higher than the original lane line recognition confidence level .
  • the above technical process is extended to the real-time scene, that is, the scene where the vehicle is driving.
  • the lane line recognition confidence level of the time stamp before the current time stamp is summed to obtain the updated lane line recognition confidence level of the current time stamp, so that the credibility/accuracy of the lane line recognition confidence level can be improved in real time.
  • the summation includes at least one of direct summation and weighted summation; in weighted summation, different weights can be assigned to different timestamps, and the distance segments (corresponding to Timestamps) farther timestamps have smaller weights and timestamps closer to the segment have larger weights.
  • weights and distances can use conventional functions such as linear or exponential functions, which are calculated by weighting. And, the updated lane line recognition confidence can be obtained more accurately.
  • further constructing the true value of the lane line based on the obtained updated lane line recognition confidence level includes: judging whether the updated lane line recognition confidence level is greater than the first threshold, and whether the lane line recognition length is greater than the second Threshold, acquiring a first time stamp set, the lane line recognition confidence level corresponding to each time stamp in the first time stamp set is greater than the first threshold, and the corresponding lane line recognition length is greater than the first threshold; For each time stamp in the time stamp set, obtain the first lane line point set from N to M meters near the vehicle in the vehicle coordinate system.
  • the updated second time stamp set with the lane line recognition confidence greater than the first threshold and the lane line recognition length less than the second threshold; for each time stamp in the second time stamp set, other timestamps that meet the set conditions are used Instead, the lane line identification length corresponding to the other timestamps is greater than the first threshold; for each timestamp in the second set of timestamps after the replacement, the distance between N and M meters at the near end of the vehicle in the vehicle coordinate system is obtained.
  • the second lane line point set The first lane line point set and the second lane line point set are clustered and grouped.
  • the true value of the lane line can be constructed based on the updated (more accurate/higher reliability) lane line recognition confidence. Therefore, compared with the original lane line recognition result, the true value of the lane line obtained through the above-mentioned technical steps will be closer to the actual situation, and is suitable as an object/reference for comparison with the lane line recognition result.
  • the abnormal event includes: the lane line recognition result is too short, the lateral error of the lane line recognition result is too large, and the lane line recognition result The heading error is too large, and the lane line recognition result is missed.
  • these abnormal events can be used as high-value training materials to be fed back to the lane line recognition algorithm located on the server or (for example, the cloud), and the recognition of the lane line recognition algorithm can be improved.
  • the recognition accuracy of the lane line recognition algorithm is improved, the lane line recognition algorithm on the vehicle side can be updated through communication. This cycle can form a benign data-training material-algorithm performance improvement cycle.
  • Various implementations of the first aspect provide an updated lane line recognition confidence that is determined based on a posteriori data.
  • the updated lane line recognition confidence is more reliable than the original lane line recognition confidence. Therefore, it can be used as the data basis for the verification and comparison of the lane line recognition results; the various embodiments of the first aspect do not need to use other data sources (such as lidar, GPS, high-definition maps, etc.), and only need to check the existing vehicles.
  • the obtained lane line recognition data and inertial navigation data are reprocessed to obtain more reliable lane line recognition results, and the obtained more reliable lane line recognition results are used to verify the original lane line recognition results.
  • the first The solution of the aspect requires a small amount of calculation, high robustness, and can be easily completed on the vehicle end; in summary, the various implementations of the first aspect provide a low-cost and highly reliable optimization solution for lane line recognition results .
  • an embodiment of the present application provides a lane line recognition device, including: a lane line recognition module provided at the vehicle end; an inertial navigation module provided at the vehicle end; A computing device connected to the module in communication; the computing device is configured to execute various technical solutions of the first aspect.
  • the embodiment of the second aspect provides a lane line recognition device built on the vehicle end, which can complete various technical solutions of the first and second aspects on the vehicle end.
  • an embodiment of the present application provides a lane line recognition system, including: the lane line recognition device of the second aspect, a lane line recognition algorithm module set on the server (the server may be a cloud or a server),
  • the lane line algorithm recognition module includes a lane line recognition algorithm; the lane line recognition device is in communication connection with the lane line recognition algorithm module, and the lane line recognition device is configured to send an abnormal event to the lane line recognition algorithm module, and the lane line
  • the recognition algorithm module is configured to use the abnormal event to train the lane line recognition algorithm.
  • the embodiment of the third aspect provides a complete system including the vehicle end and the service end, which can complete various technical solutions of the first to second aspects.
  • an embodiment of the present application provides a computer-readable storage medium, including an instruction set, which can be executed by a processor to implement various technical solutions of the first aspect.
  • an embodiment of the present application provides a lane line recognition system, which includes a processor and a computer-readable storage medium.
  • the computer-readable storage medium stores an instruction set that can be executed by the processor to implement the first aspect described above.
  • Various embodiments of the present application provide a method for determining the confidence level of lane line recognition, a lane line recognition device, a lane line recognition system, a non-transient storage medium, and a lane line recognition system.
  • the various technical solutions of this application obtain updated and more reliable lane line recognition confidence based on the existing posterior data, and the lane line truth obtained based on the updated and more reliable lane line recognition confidence.
  • the value is compared with the (original) lane line recognition result as a reference to obtain lane line recognition abnormal events, and the lane line recognition abnormal events can be fed back to the lane line recognition algorithm as high-value training materials to train the lane line recognition algorithm.
  • the recognition accuracy of the lane line recognition algorithm is improved, and the lane line recognition algorithm that has improved the recognition accuracy can be sent to the car end in one step, thus forming a virtuous circle.
  • the technical solution of this application does not need to use other data sources (such as lidar, GPS, high-definition maps, etc.), and only needs to reprocess the lane line recognition data and inertial navigation data that have been obtained by the existing vehicles to obtain a more reliable Lane line recognition results, and the amount of calculation required by the technical solution of this application is small, and the entire verification and acquisition of abnormal events can be completed on the vehicle side. Therefore, the technical solution of this application has the technical advantages of low cost and high reliability.
  • the driving field has broad application prospects.
  • FIG. 1 is a schematic diagram of a method for determining an abnormal event of lane line recognition according to an embodiment of the present application
  • FIG. 2 is a schematic diagram of determining the maximum lane line recognition length and the first time interval in the determination of the lane line recognition confidence level provided by an embodiment of the present application;
  • FIG. 3 is a schematic diagram of the calculation process of the updated lane line recognition confidence provided by the embodiment of the present application.
  • FIG. 4 is a schematic diagram of the process of obtaining the updated lane line recognition confidence level provided by an embodiment of the present application
  • FIG. 5 is a schematic diagram of the true value construction of lane lines provided by an embodiment of the present application.
  • Fig. 6 is a schematic diagram of a lane line recognition device and a lane line recognition system provided by an embodiment of the present application.
  • FIG. 7 is a schematic diagram of a computer-readable storage medium and a processor provided by an embodiment of the present application.
  • Fig. 1 shows the overall flow chart of the method for identifying abnormal events based on the lane line recognition according to the embodiment of the present application, which mainly includes:
  • S1 Data preprocessing.
  • the data acquired by the vehicle module (the acquired data includes lane line recognition results and inertial navigation data) are preprocessed, including: various data sources are aligned based on the timestamp;
  • S2 The confidence of the updated lane line recognition is determined. Based on the aligned data of S0, obtain the updated lane line recognition confidence level according to the set logic;
  • S3 Lane line truth value construction. Based on the updated lane line confidence obtained by S2, construct the true value of the lane line;
  • S4 Lane line recognition abnormal event extraction. Based on the lane line recognition result and the true value of the lane line obtained by S3, extract the lane line recognition results of multiple abnormal types;
  • the vehicle-based automatic driving system obtains the lane line recognition result data and inertial navigation data, and aligns the lane line recognition result with the inertial navigation data in time.
  • the lane line recognition result is based on the lane line recognition at the vehicle end
  • the lane line recognition module obtains the lane line recognition result through the image information obtained by the vehicle and related algorithms (such as but not limited to IMM: Interacting Multiple Model).
  • the inertial navigation data is obtained based on the inertial navigation module. Based on the inertial navigation data, a relatively accurate spatial positioning of the vehicle within a short distance range (for example, within 100 meters) can be obtained.
  • the update frequency of the lane line recognition module is 15 Hz, and the update frequency of the inertial navigation module is 100 Hz.
  • the acquired inertial navigation data is aligned based on the lane line recognition data, that is, the aligned lane line recognition result data
  • the aligned multiple time stamps constitute a time stamp sequence, and each time stamp in the time stamp sequence represents a specific time.
  • any other suitable frequency and alignment standard can be based on Depending on the actual scene, by aligning the lane line recognition result with the inertial navigation data, the spatial position of the vehicle and the lane line recognition result at different moments can be determined based on the time axis.
  • the step of determining the confidence of the updated lane line recognition can be performed.
  • the updated lane line recognition confidence level is determined based on the posterior data (that is, the data acquired after the vehicle is driven).
  • the posterior data can include the time-aligned lane line recognition result data and inertial navigation data described in step S1, so it can also be called offline data.
  • the lane line recognition confidence can be performed based on the posterior data. Degree of determination.
  • the vehicle speed at the time of a certain time stamp can be multiplied by the time interval between time stamps to achieve this purpose.
  • the vehicle speed and direction at time T2 may be different from the vehicle speed and direction at time T1, since the time interval between adjacent time stamps is 0.066 seconds, this time range is different for general vehicles.
  • the speed/direction change of the vehicle caused by acceleration/deceleration or direction change is small and negligible, so the above process can be calculated as equivalent to a uniform linear motion.
  • the speed of the vehicle does not exceed 120km/h, so it can be considered that the maximum displacement S MAX of the vehicle between two time stamps is about 30 meters/second*0.066 ⁇ 2 meters.
  • the lane line recognition result data includes lane line recognition length X and lane line recognition confidence level C. It should be understood that the above-mentioned lane line recognition length and lane line recognition confidence level are the same as the time obtained in step S1.
  • the data corresponding to the stamp sequence that is, for each time stamp in the time stamp sequence acquired in S1, it has a lane line recognition length corresponding to it and a lane line recognition confidence corresponding to it Spend.
  • the lane line identification lengths corresponding to different time stamps may be different.
  • the identification length of the lane line at time stamp T1 (time 0) is 20 meters
  • the identification length of the lane line at time stamp T2 (time 0.066 seconds) immediately adjacent to T1 is 35 meters.
  • the lane line recognition confidence level C can take a value in the range of zero to one. Zero represents that the lane line recognition result is completely unreliable, and one represents that the lane line recognition result is completely credible.
  • Each lane line recognition confidence C has a corresponding confidence interval [N, M], where: the value range of N is 4-6 meters, and the value range of M is 8-12 meters.
  • N the value range of N is 4-6 meters
  • M the value range of M is 8-12 meters.
  • X is greater than M; but in some special cases, such as missing lane lines or intersection scenarios, the lane line recognition length X may also be less than M.
  • the confidence interval [N, M] means that at the time of the current timestamp, the recognition range of the vehicle within the range of N to M meters in front of the vehicle in the direction of travel of the lane on which it is located is “reliable” ( It should be understood that: “reliable” here refers to: Compared with other spatial intervals of lane line recognition length X, the lane line recognition confidence result in the confidence interval [N,M] is more accurate and credible ; “Reliable” here does not mean that the value of the lane line recognition confidence in the confidence interval [N, M] is close to one [that is, the lane line recognition result is completely credible]); in other words, at the current timestamp At the moment, for the acquired lane line recognition range X, the lane line recognition confidence level within the confidence interval [N, M] is relatively reliable;
  • the above-mentioned confidence interval [N, M] changes according to the change of the lane, that is, the value range of the confidence interval of the vehicle's lane and the confidence interval of the adjacent lane is Different, for example, in some embodiments, for the current lane where the vehicle is located, the value range of N is 3-6 meters, and the value range of M is 8-12 meters; for the adjacent lanes of the current lane, The value range of N can be 9-11 meters, and the value range of M can be 13-17 meters.
  • the maximum lane line recognition length L is introduced, and L represents the maximum lane line recognition length set a priori, that is, at any time, the length of the lane line recognized by the vehicle will not be greater than L, and L can take 50- Value within 100 meters. In some embodiments, L can take 70 meters.
  • each time stamp after data alignment has a lane line recognition length X and a lane line recognition confidence level C corresponding to it.
  • the vehicle has a specific and determined vehicle position corresponding to it in space, and once the coordinate system of the vehicle is determined, the vehicle position can be used The specific coordinates are expressed in the form.
  • the aforementioned offline and posterior data are processed one by one according to the timestamp to obtain the updated lane line recognition confidence.
  • T timestamp as an example to describe in detail the specific steps for obtaining the updated lane line recognition confidence:
  • Fig. 2 also shows a schematic diagram of the lane line recognition confidence interval N to M at the time of the T time stamp.
  • N may be 4 meters, and M may be 9 meters.
  • the specific process of obtaining the first time interval [S1, E1] includes:
  • the recognition confidence interval is in the current coordinate system (T
  • T The time vehicle coordinate system
  • N and M take 4 and 9 (meters) respectively
  • the above range of [0, MN] completely covers the range of [0, N]
  • the lane line recognition confidence level of the timestamp corresponding to the vehicle position in the negative direction N of the Y axis can be used to "replace" the lane line recognition confidence level of the vehicle in the range of 0 to N in front of the time T.
  • the confidence in the range from L-N to L at time T can be replaced by the lane line recognition confidence of the time stamp corresponding to the vehicle position at the far end L-M;
  • the first time interval [S1, E1] is obtained based on the time stamps S1 and E1. It can be seen that the vehicle space position interval corresponding to the first time interval is [-N ⁇ S,L-M ⁇ S].
  • the line recognition confidence level Ci it can also determine the coordinate position of the vehicle at the time Pi; and the lane line recognition confidence level at the time Pi is the most reliable at the front of the driving direction of the vehicle location [N,M].
  • the position corresponding to S2 is: the position N ⁇ S below the current position of the vehicle (the negative direction of the Y axis/the opposite direction of the vehicle traveling direction), and the position corresponding to E2
  • the position is the position of Xi-M ⁇ S above the current position of the vehicle (positive direction of the Y axis); the space range between the positions corresponding to S2 and E2 is the space corresponding to the time interval [S2, E2] distance.
  • the time stamp sequence included/corresponding to the lane line recognition range is recorded as the sequence Ki: ⁇ k i1 , K i2 ,,...k im ⁇ .
  • the elements in the K2 and K4 sequences corresponding to them may include coincident timestamps, that is to say For one or several timestamps in the K2 and K4 sequences, they may appear either in the second time sequence K2 corresponding to P2 or in the second time sequence K4 corresponding to P4.
  • the lane line recognition confidence level of the second time series corresponding to the P-sequence time stamp is summed to obtain the updated lane line recognition confidence level of the time stamp; for example, if a time stamp is in the first time corresponding to P1, P3, and P4 in total If there are three occurrences in the second time series, then the updated lane line recognition confidence level of the timestamp is the sum of the lane line recognition confidence levels of P1, P3, and P4. It should be understood that the summation here can be either a direct summation or a weighted summation.
  • the updated lane line recognition confidence obtained after the summation may be divided by the number of sums to achieve normalization.
  • Ki sequence and P sequence are both constructed based on the same set of time stamp data, that is, after data alignment is performed in step S1 Timestamp data.
  • Fig. 4 shows Xi, Xj, Xk corresponding to Pi, Pj, Pk respectively; in the exemplary illustration of Fig. 4, Xi, Xj, Xk are all greater than M; Fig. 4 also shows a time stamp Pt, and the lane line identification length Xt corresponding to Pt is less than M;
  • the second time interval corresponding to them can be obtained respectively, and the distance interval corresponding to each second time interval is Li, Lj, Lk as shown in the figure.
  • the distance interval corresponding to each second time interval is Li, Lj, Lk as shown in the figure.
  • N ⁇ S and Xk-M ⁇ S are not directly shown in Figure 4, but the current The [-N,Xk-M] interval where the vehicle position is the reference point is used as an indication. After the Li space position is determined, the second time interval corresponding to Xi can be determined according to the correspondence between time and space;
  • Li and Lj can also be obtained respectively.
  • the corresponding second time intervals can be obtained respectively, and then the timestamps included in each second time interval can be determined, as can be seen
  • the three timestamps of Pi, Pj, and Pk they are all included in the three intervals Li, Lj, and Lk. Therefore, the updated lane line of any one of the three timestamps of Pi, Pj, and Pk is calculated When identifying the confidence, the confidence of the lane line recognition at the three moments of Pi, Pj, and Pk must be summed.
  • the confidence level of different segments of the lane line can be determined spatially.
  • the updated lane line recognition confidence is equivalent to the lane line recognition confidence in the range of N to M meters in front of the vehicle position at the time of the time stamp; it should be understood that:
  • the space segments corresponding to the stamps may or may not overlap;
  • the updated lane line recognition confidence is calculated and obtained.
  • the credibility/accuracy of the lane line recognition confidence of each time stamp can be improved. Because for a certain time stamp, there is only one value of the lane line recognition confidence level corresponding to it in the prior art, and the accuracy of the lane line recognition confidence level is greatly affected by the environment, for example, if it is in a certain time stamp At the moment, when the vehicle is driving with strong light in the opposite direction, and the strong light affects the image acquisition of the front environment by the vehicle’s camera, it is very likely that the result of the lane line recognition confidence at the time stamp will be greater.
  • the technical solution in the embodiment of the present application includes its time stamp (which can be recorded/understood as the time “corresponding” to any time stamp).
  • the sum of the lane line recognition confidence of the stamp set is used to determine the updated lane line recognition confidence of the timestamp, so that the error of the lane line recognition confidence caused by the recognition fluctuation of a single timestamp is eliminated to the maximum extent/ inhibition.
  • the updated lane line recognition confidence level reflects all possible lane line recognition lengths as a whole (or the lane line recognition confidence interval).
  • the second time interval corresponding to the length) includes the confidence level of the time stamp of the lane line recognition confidence interval; thus, the technical solution of the embodiment can suppress/eliminate the inaccuracy of the lane line recognition confidence level caused by the recognition fluctuation ; Improve the credibility/accuracy of the lane line recognition confidence.
  • two, three, or several recognition ranges that are closest to the timestamp can be selected and the lane line recognition confidence levels including its timestamp can be summed to obtain
  • the updated lane line recognition confidence level of the timestamp can also be selected from the timestamps whose identification range includes the timestamp with the longest identification range to determine the updated lane line of the timestamp Identify the confidence level; those skilled in the art can make a selection based on actual needs without departing from the spirit of this application.
  • Figure 4 exemplarily summarizes the above technical process 400, which specifically includes:
  • the above technical process can be extended to a real-time scene, that is, a scene where the vehicle is driving.
  • the lane line recognition confidence level of the time stamp may be two, three or more) before the current time stamp is summed to obtain the updated lane line recognition confidence level of the current time stamp, so that the lane can be improved in real time
  • the degree of credibility/accuracy of the line recognition confidence may be two, three or more
  • S3 is the lane line truth value construction step.
  • the purpose of this step is to construct the lane line truth value based on the updated lane line recognition confidence obtained by S2.
  • the first time interval [S1,E1] of T timestamp and all the timestamps in the first time interval ⁇ p 1 , p 2 ,...p n ⁇ are obtained in S2
  • the updated lane line recognition confidence level is obtained.
  • FIG. 5 it exemplarily shows the flow 500 of lane line truth value construction, including:
  • the threshold can be set to 0.6. It should be understood that the technician can set the threshold according to actual needs Does not deviate from the spirit of this application;
  • step 503 Determine whether the lane line identification length corresponding to the timestamp is greater than M. If the lane line identification length is less than M, it will enter the offline lane line tracking process 508 (shown as a dashed box in Figure 5); if the lane line identification length is greater than M , Then go to step 504;
  • the process 508 includes:
  • the setting coefficient can be selected from a value in the range of 0.4 to 0.6;
  • step 5083 Determine whether the value obtained in step 5082 is greater than M, if it is greater than M, go to step 504, otherwise, return to the previous step to continue traversing the next time stamp;
  • Clustering and grouping the obtained lane line point set you can use the conventional clustering method to cluster the lane line point set, for example, you can use DBSCAB (Denisty-Based Spatial Clustering of Application with Noise) Method, it should be understood that any other suitable clustering method can also be used;
  • DBSCAB Dynamic-Based Spatial Clustering of Application with Noise
  • the collinear connection judgment includes two steps. The first step: the interval The lane lines on both sides are vectorized, and the angle or horizontal distance between the two vectors is judged.
  • the angle exceeds the set threshold (for example, 30 degrees), or the horizontal distance is greater than the set threshold (for example, 3 meters), then judge There is no need to make collinear connections; and if it is judged that the angle and lateral distance between the two vectors are both smaller than the set threshold, enter the second step, which is to consider the lane line recognition confidence of the timestamp within the interval.
  • the lane line recognition confidence of most time stamps in the range is less than the set threshold (for example, the above 0.6), then the collinear connection is still not performed, and when the lane line recognition confidence of most time stamps in the interval is greater than the set threshold When, the collinear connection is carried out;
  • step 507 polynomial fitting is performed on the lane line obtained in 506, and the true value of the lane line at the time T timestamp is obtained.
  • the comparison objects include: the true value of the lane line obtained through S3 and the lane line recognition result obtained by the vehicle.
  • the lane line recognition result can include The lane line recognition algorithm on the cloud is based on the image information obtained by the vehicle.
  • the evaluation can be based on preset indicators.
  • the preset indicators may include: (1) The lane line recognition result is too short; (2) The lane line recognition result is too large in lateral error; (3) The lane line recognition result is too large in orientation ; (4) Missing detection of lane line recognition results.
  • a polynomial may be used to fit the lane line.
  • the true length of the lane line as Ym
  • the identification length of the lane line as Yc (meters)
  • the truncation point as x 1 , x 2 , ... x n
  • the truncation point refers to the lane
  • the preset points on the trajectory of the line recognition result such as points 10 meters, 20 meters, and 30 meters away from the origin, those skilled in the art can specifically select a suitable cutoff point according to the actual situation.
  • the evaluation process and meaning of the preset indicators are as follows:
  • the set threshold here can be 0.5. It should be understood that those skilled in the art can use the actual conditions to determine Choose an appropriate threshold without departing from the spirit of this application;
  • the set threshold may be 1 meter. It should be understood that those skilled in the art can choose according to the actual situation. Appropriate threshold without departing from the spirit of this application;
  • Missing detection of lane line recognition result there is a case where the true value of the lane line Ym is present but the lane line recognition result Ym is not output.
  • the lane line recognition algorithm may be set on the server side, and the abnormal lane line recognition data may be transmitted to the lane line recognition algorithm on the server side to improve the recognition accuracy of the algorithm.
  • the server can be a cloud, and in some embodiments, the server can also be a server.
  • the lane line recognition algorithm in the cloud can also provide more accurate/reliable lane line recognition capabilities for the car terminal through the communication network after training. Such a cycle can form a benign data-training material-algorithm performance improvement cycle.
  • the lane line recognition device includes a lane line recognition module, an inertial navigation module, a computing device, which are arranged on the vehicle side, and the computing device communicates with the lane line recognition module and the inertial navigation module. It is connected and configured to execute the solutions of the various embodiments of S0-S6 described above.
  • the system includes a lane line recognition device set on the vehicle side and a lane line recognition algorithm module set on the server side; the lane line recognition device includes a lane line recognition module and an inertial navigation Module, computing device; the lane line recognition device is in communication connection with the lane line recognition algorithm module, the lane line recognition device is configured to send lane line recognition abnormal events to the lane line recognition algorithm module, the lane line recognition algorithm module It is configured to use lane line recognition abnormal events to train the lane line recognition algorithm to optimize the recognition accuracy of the lane line recognition algorithm.
  • the server can be a cloud, and in some embodiments, the server can also be a server.
  • the computer-readable storage medium 71 includes an instruction set 73, which can be executed by a processor to implement various implementations of S0-S6.
  • the lane line recognition system includes a processor 72 and a computer-readable storage medium 71.
  • the computer-readable storage medium stores an instruction set 73, which can be used by the processor. Execute to realize the various implementations of S0-S6 described above.
  • Various embodiments of the present application provide a method and system for detecting a vehicle's drivable area, and an autonomous vehicle adopting the system.
  • multiple sensors are used, and the probability information of obstacles obtained based on multiple sensors is combined to obtain a probability grid map, so that the drivable area around the vehicle can be more accurately identified; on the other hand,
  • This application uses obstacle parallax information, obstacle mapping, etc., so that the technical solution of this application can have high generalizability, is widely applicable to a variety of scenarios, and does not rely on training data. Therefore, the technical solution of this application Robust.
  • the technical solution of this application can be widely applied to different levels of automatic driving solutions, systems, and vehicles.
  • the disclosed system, device, and method can be implemented in other ways.
  • the device embodiments described above are merely illustrative, for example, the division of units is only a logical business division, and there may be other divisions in actual implementation, for example, multiple units or components can be combined or integrated. To another system, or some features can be ignored, or not implemented.
  • the displayed or discussed mutual coupling or direct coupling or communication connection may be indirect coupling or communication connection through some interfaces, devices or units, and may be in electrical, mechanical or other forms.
  • the units described as separate components may or may not be physically separate, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
  • business units in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
  • the above-mentioned integrated unit can be realized in the form of hardware or software business unit.
  • the integrated unit is implemented in the form of a software business unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
  • the technical solution of the present application essentially or the part that contributes to the existing technology or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium , Including several instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute all or part of the steps of the methods in the various embodiments of the present application.
  • the aforementioned storage media include: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), magnetic disks or optical disks and other media that can store program codes. .
  • the services described in this application can be implemented by hardware, software, firmware, or any combination thereof.
  • these services can be stored in a computer-readable medium or transmitted as one or more instructions or codes on the computer-readable medium.
  • the computer-readable medium includes a computer storage medium and a communication medium, where the communication medium includes any medium that facilitates the transfer of a computer program from one place to another.
  • the storage medium may be any available medium that can be accessed by a general-purpose or special-purpose computer.

Abstract

一种车道线识别置异常事件确定方法、车道线识别装置及系统，方法包括：依据后验的车道线识别结果确定更新的车道线识别置信度；依据更新的车道线识别置信度构建车道线真值；依据车道线识别结果和车道线真值确定车道线识别异常事件。依据后验的车道线识别结果来更新之前所获取的车道线识别置信度，提升了车道线识别置信度的准确性和可靠性；并利用更新的车道线识别置信度来构建车道线真值，通过构建的车道线真值和车道线识别置信度可以获取车道线识别异常事件，这些异常事件可以用来进一步训练车道线识别算法，从而提升车道线识别算法的识别准确性和精度。

Description

车道线识别异常事件确定方法、车道线识别装置及系统 技术领域
本申请涉及自动驾驶领域，特别地，涉及一种车道线识别异常事件确定方法、车道线识别装置及系统。
背景技术
随着5G通信和车联网技术的快速发展，自动驾驶技术已经成为研究热点。自动驾驶领域核心技术包括智能环境感知、自动导航定位、驾驶行为决策和智能路径规划控制等。在自动驾驶过程中，车辆需要对道路中的车道线进行检测，然后车辆基于所检测到的车道线以及开启的功能，完成驾驶行为。车道线识别结果一般由设置在车端的车道线识别装置基于算法对车辆在行驶过程中所获取的图像信息进行处理而得到，目前，算法往往采用基于神经网络的深度学习方案(例如但不限于CNN(卷积神经网络)等)。
虽然目前深度学习在图像识别等领域取得了很大的进展，但是，对于车道线识别而言，目前的识别结果仍无法100％可靠，存在的常见问题包括车道线漏检和误检。而对于车道线识别算法而言，训练材料又是非常重要的，如果不能获得漏检和误检的事件，算法将难以提升识别精度，因需要对车道线识别结果进行校验，获取车道线识别异常事件，并基于这些有效异常数据对车道线识别算法进行优化(算法的正向反馈)，进而评价新算法是否较原算法更优，促进车道线识别算法迭代更新。因此，对于车道线识别结果进行校验必然要求使用另外一套参照体系对比已有的基于算法获得的车道线识别结果进行比较才可以实现。
目前，对车道线识别结果进行校验最直接的方法是依赖人工进行校验，但是采用人工校验车道线识别结果的正确与否，效率低下并且耗费极高。另一类方案通过直接下载高精度地图，基于高精地图上的车道线信息和车辆的车道线识别结果进行对比。但是：1.高精度地图制作成本高，导致可用范围有限；2.可能存在更新延迟的问题；3.在被高层建筑、立交桥、隧道等地物遮挡GPS的路段，会影响卫星信号的接收，从而可能导致系统出现定位数据漂移现象，严重时甚至可能失去信号。还有一类方案通过激光雷达方法来识别车道线，但如果采用激光雷达类方法，则受车道线反射率的限制，如果激光点云中车道线的反射率较低，则会影响车道线的生成精度；如果增加激光点云的数据量，则需要大量存储空间且消耗大量的运算时间，使得激光雷达类方法在车端应用成本高昂。
综上，可以看出：目前对于车道线识别结果进行校验缺乏可靠且价格低廉的解决方案，因而对于车道线识别算法而言也由于异常数据的难以获得而精度提升困难。
发明内容
为了解决上述问题，本申请各种实施例提供了一种车道线识别置信度确定方法、一种车道线识别装置、一种车道线识别系统以及一种计算机可读存储介质。
一方面，本申请实施例提供了一种车道线识别置信度确定方法，该方法具体可以包括： 首先依据后验的车道线识别结果确定更新的车道线识别置信度；然后基于更新的车道线识别置信度构建车道线真值；最后基于车道线识别结果和所述车道线真值确定车道线识别异常事件，所述后验的车道线识别结果包括当车辆行驶完成后所获取的车道线识别置信度。本申请依据后验的车道线识别结果来更新之前所获取的车道线识别置信度，提升了车道线识别置信度的准确性和可靠性；并利用更新的车道线识别置信度来构建车道线真值，通过构建的车道线真值和车道线识别置信度可以获取车道线识别异常事件，这些异常事件可以用来进一步训练车道线识别算法，从而提升车道线识别算法的识别准确性和精度。
在一个可能的设计中，将所述后验的车道线识别结果和后验的惯性导航数据基于时间戳进行对准；所述后验的惯性导航数据包括当车辆行驶完成后所获取的惯性导航数据。，通过将车道线识别结果和惯性导航数据进行时间对准，可以基于时间轴确定在不同时刻车辆的空间位置和车道线识别结果。车道线识别结果包括：车辆位置、车道线识别置信度和车道线识别长度。
在一个可能的设计中，对于任一时间戳，至少基于与所述时间戳对应的所述后验的车道线识别结果中的两个车道线识别置信度确定与所述时间戳对应的更新的车道线识别置信度。技术步骤可以包括：对于任一时间戳，确定与所述任一时间戳对应的时间戳集合，所述时间戳集合中包括一个或多个时间戳，所述一个或多个时间戳的车道线识别长度均包括与所述任一时间戳对应的车辆位置，，获取所述时间戳集合中每个时间戳对应的车道线识别置信度并构成车道线识别置信度集合，从所述车道线识别置信度集合中获取至少两个进行求和以得到与所述任一时间戳对应的更新的车道线识别置信度。或者，在另一些情况下，技术步骤可以包括：对所述车道线识别置信度集合中的所有车道线识别置信度进行求和以得到与所述任一时间戳对应的更新的车道线识别置信度。对于任一时间戳而言，通过对至少两个/多个车道线识别范围包括它的时间戳的车道线识别置信度的求和来确定该时间戳的更新的车道线识别置信度，使得由单一时间戳的识别波动所导致的车道线识别置信度的误差被抑制，从而对于任一时间戳而言，更新的车道线识别置信度的精度/可靠性要高于原始的车道线识别置信度。
在一个可能的设计中，将上述技术过程延伸到实时场景，即车辆正在行驶的场景，在此种情况下，可以选取例如车道线识别范围包括当前时间戳的多个前序的(时间上在当前时间戳之前的)时间戳的车道线识别置信度求和来获取当前时间戳的更新的车道线识别置信度，从而可以实时地提升车道线识别置信度的可信程度/准确性。
在一个可能的设计中，所述求和包括直接求和与加权求和中的至少一个；在加权求和中，可以采用将不同的权重赋予不同时间戳，在时间上距离分段(所对应的时间戳)较远的时间戳具有较小的权重而距离分段较近的时间戳具有较大的权重，上述权重和距离的关系可以采用线性或者指数函数等习知的函数，通过加权求和，可以更加准确地获取更新的车道线识别置信度。
在一个可能的设计中，进一步地依据获取的更新的车道线识别置信度构建车道线真值，包括：判断更新的车道线识别置信度是否大于第一阈值，以及车道线识别长度是否大于第二阈值，获取第一时间戳集合，所述第一时间戳集合中的每个时间戳对应的车道线识别置信度大于第一阈值，对应的车道线识别长度大于第一阈值；对所述第一时间戳集合中的每个时间戳，获取车辆坐标系下车辆近端N至M米的第一车道线点集。获取更新的车道线识别置信度大于第一阈值且车道线识别长度小于第二阈值的第二时间戳集合；对于第二时间戳集合中的每个时间戳，以符合设定条件的其它时间戳来代替它，所述其他时间戳对应的车道线识别长 度大于第一阈值；对代替后的第二时间戳集合中的每个时间戳，均获取车辆坐标系下车辆近端N至M米的第二车道线点集。对第一车道线点集和第二车道线点集进行聚类分组，如果分组后车道线点集组之间的纵向距离大于距离阈值，则基于车道线存在性结果的判断决定是否进行共线连接。通过上述过程，可以基于更新的(更准确的/可靠性更高的)车道线识别置信度来构建车道线真值。因此较之原始的车道线识别结果，通过上述技术步骤所获得的车道线真值将更加接近实际情况，适宜于作为与车道线识别结果进行比较的对象/参照。
在一个可能的设计中，比较获取的车道线真值和车道线识别结果以确定异常事件，所述异常事件包括：车道线识别结果过短、车道线识别结果横向误差过大、车道线识别结果朝向误差过大，车道线识别结果漏检。
在一个可能的设计中，在获取了异常事件后，，这些异常事件可以作为高价值的训练材料反馈给位于服务端或(例如云端)的车道线识别算法，并可提升车道线识别算法的识别准确率，而当车道线识别算法的识别准确率提升后，又可以通过通信方式更新车端的车道线识别算法，如此循环，可以形成良性的数据-训练材料-算法性能提升的循环。
第一方面的各种实施方式提供了基于后验数据而确定的更新的车道线识别置信度，更新的车道线识别置信度较之原始的车道线识别置信度而言具有更高的可靠性，因此可以作为对车道线识别结果进行校验比较的数据基础；第一方面的各种实施例无需使用其它的数据源(例如激光雷达、GPS、高清地图等)，只需对已有的车辆已经获得的车道线识别数据和惯性导航数据进行再加工即可获得更加可靠的车道线识别结果，并利用获得的更加可靠的车道线识别结果来对原始车道线识别结果进行校验，最后，第一方面的方案的所需的计算量小，鲁棒性高，可以轻易地在车端完成；综上，第一方面的各种实施方式提供了成本低廉、可靠性高的车道线识别结果优化方案。
第二方面，本申请实施例提供一种车道线识别装置，包括：设置于车端的车道线识别模块；设置于车端的惯性导航模块；设置于车端并与所述车道线识别模块和惯性导航模块通信连接的计算装置；计算装置被配置为可执行第一方面的各种技术方案。第二方面的实施例提供了一种架构在车端的车道线识别装置，其可以在车端完成第一和第二方面的各种技术方案。
第三方面，本申请实施例提供一种车道线识别系统，包括：第二方面的车道线识别装置，设置在服务端(服务端可以是云端，也可以是服务器)的车道线识别算法模块，车道线算法识别模块包括车道线识别算法；车道线识别装置和所述车道线识别算法模块通信连接，车道线识别装置被配置为将异常事件发送给所述车道线识别算法模块，所述车道线识别算法模块被配置为可使用所述异常事件对所述车道线识别算法进行训练。第三方面的实施例提供了完整的包括车端和服务端的系统，其可以完成第一至第二方面的各种技术方案。
第四方面，本申请实施例提供一种计算机可读存储介质，包括指令集，所述指令集可以被处理器执行以实现第一方面的各种技术方案。
第五方面，本申请实施例提供一种车道线识别系统，其包括处理器和计算机可读存储介质，计算机可读存储介质存储有指令集，指令集可以被处理器执行以实现上述第一方面的各种技术方案。
本申请的各种实施例提供了一种车道线识别置信度确定方法，一种车道线识别装置、一种车道线识别系统、一种非暂态存储介质以及一种车道线识别系统。本申请的各种技术方案基于已有的后验数据获取更新的、可靠性更高的车道线识别置信度，并基于更新的、可靠性更高的车道线识别置信度所获得的车道线真值作为基准和(原始的)车道线识别结果进行比 较，从而得到车道线识别异常事件，并可以车道线识别异常事件作为高价值训练材料反馈给车道线识别算法来训练车道线识别算法，从而可以提升车道线识别算法的识别准确度，而提升了识别准确度的车道线识别算法又可以一步发送到车端，从而形成良性循环。本申请的技术方案无需使用其它的数据源(例如激光雷达、GPS、高清地图等)，只需对已有的车辆已经获得的车道线识别数据和惯性导航数据进行再加工即可获得更加可靠的车道线识别结果，并且本申请技术方案所需要的计算量小，整个校验和异常事件的获取都可以在车端完成，因此本申请技术方案具有成本低廉、可靠性高的技术优势，在自动驾驶领域具有广阔的应用前景。
附图说明
图1是本申请实施例提供的一种车道线识别异常事件确定方法示意图；
图2是本申请实施例提供的车道线识别置信度确定中的最大车道线识别长度和第一时间区间的确定示意图；
图3是本申请实施例提供的更新的车道线识别置信度的计算过程示意图；
图4是本申请实施例提供的获取更新的车道线识别置信度流程示意图；
图5是本申请实施例提供的车道线真值构建示意图；
图6是本申请实施例提供的车道线识别装置和车道线识别系统的示意图。
图7是本申请实施例提供的计算机可读存储介质和处理器的示意图；
具体实施方式
参见图1，其示出了基于本申请实施例给出的车道线识别异常事件确定方法的整体流程图，主要包括：
S0：开始；
S1：数据预处理。在S1中，对车载模块获取的数据(获取的数据包括车道线识别结果和惯性导航数据)进行预处理操作，包括：各类数据源基于时间戳进行对准；
S2：更新的车道线识别置信度确定。基于S0的对准的数据，按照设定逻辑获取更新的车道线识别置信度；
S3:车道线真值构建。基于S2得到的更新的车道线置信度，构建车道线真值；
S4：车道线识别异常事件提取。基于车道线识别结果和S3获取的车道线真值，提取多种异常类型车道线识别结果；
S5:异常事件上传云端信息平台。将车端搜集到的异常数据上传云端信息平台，在云端进行车道线识别算法的再训练、验证和更新；
S6:结束。
下面将结合附图对上述的各个步骤进行详细阐述。
S0:开始
S1：数据预处理
参见图6，基于车端的自动驾驶系统获取车道线识别结果数据和惯性导航数据，并将车道线识别结果和惯性导航数据进行时间对准，一般而言，车道线识别结果基于车端的车道线识别模块而得到，车道线识别模块通过车辆所获取的图像信息以及相关的算法(例如但不限于 IMM:Interacting Multiple Model)得到车道线识别结果。惯性导航数据基于惯性导航模块而得到，基于惯性导航数据可以获得车辆在短距离范围内(例如100米以内)的较为精确的空间定位。
在一些实施例中，车道线识别模块的更新频率为15Hz，惯性导航模块的更新频率为100Hz，将获取的惯性导航数据基于车道线识别数据进行对准，即对准后的车道线识别结果数据和惯性导航数据的时间戳之间的间隔均为1/15＝0.066秒。对准后的多个时间戳构成一个时间戳序列，时间戳序列中的每一个时间戳代表一个具体的时刻。可以理解的是：虽然在实施例中示出了上述的频率(15Hz,100Hz)以及将车道线识别模块的更新频率作为对准基准，但是任何其它合适的频率以及对准的标准都是可以依据实际场景而改变的，通过将车道线识别结果和惯性导航数据进行时间对准，可以基于时间轴确定在不同时刻车辆的空间位置和车道线识别结果。
S2：更新的车道线识别置信度确定
当S1步骤完成后，即可进行更新的车道线识别置信度确定的步骤。
在一些实施例中，基于后验数据(即车辆行驶完成后所获取的数据)来进行更新的车道线识别置信度的确定。后验数据可以包括上述S1步骤中所述的时间对准后的车道线识别结果数据和惯性导航数据，所以也可以称为离线数据，在S2步骤中，可以基于后验数据进行车道线识别置信度的确定。
在一些实施例中，基于惯性导航数据来计算在相邻时间戳时间范围内车辆发生的位移，可以使用某一时间戳所在时刻的车辆速度乘以时间戳之间的时间间隔来实现该目的，仍以上一段描述中的时间戳为例：若在时间戳T1(0时刻)的车速为80km/h(即V T1约22m/S)，则可以通过V T1*0.066S＝1.45m来获取时间戳T1和T2之间车辆发生的位移。应当理解的是：虽然在T2时刻的车辆速度、方向等矢量均可能和T1时刻的车辆速度、方向有不同，但是由于相邻时间戳之间时间间隔为0.066秒，这个时间范围对于一般车辆而言进行加速/减速或者方向变化所导致的车辆速度/方向变化是微小而可以忽略不计的，所以可以将上述过程等价于匀速直线运动来进行计算。一般而言，车辆行驶速度不超过120km/h，所以可以认为两个时间戳之间车辆的最大位移S MAX约为30米/秒*0.066≈2米。
在一些实施例中，车道线识别结果数据包括车道线识别长度X和车道线识别置信度C，应当理解的是，上述车道线识别长度和车道线识别置信度均是和步骤S1中获取的时间戳序列相对应的数据，即：对于S1中获取的时间戳序列中的每一个时间戳而言，其均具有一个与它相对应的车道线识别长度和一个与它相对应的车道线识别置信度。
应当理解的是：不同时间戳所对应的车道线识别长度可能是不同的。例如：在时间戳T1(0时刻)时刻的车道线识别长度为20米，而在紧邻T1的时间戳T2(0.066秒时刻)时刻的车道线识别长度为35米。
车道线识别置信度C可以取零至一范围内的数值，零代表车道线识别结果完全不可信，一代表车道线识别结果完全可信。每个车道线识别置信度C均具有与之对应的置信度区间[N,M]，其中：N的取值范围为4-6米，M的取值范围为8-12米。一般而言，对于一个时间戳而言，X大于M；但是在一些特殊情况，例如车道线缺失或者路口场景下，车道线识别长度X也可能小于M。对于一时间戳而言，置信度区间[N,M]指在当前时间戳时刻，车辆对于其所处车道的行进方向前方距离车辆N至M米的范围内的识别范围是“可靠的”(应当理解：这里的“可靠的”指的是：较之车道线识别长度X的其它空间区间而言，在置信度区间[N,M] 内的车道线识别置信度结果是更加准确可信的；这里的“可靠的”并非指在置信度区间[N,M]内的车道线识别置信度的数值是接近于一[即车道线识别结果完全可信]的)；换言之，在当前时间戳时刻，对于所获取的车道线识别范围X而言，在置信度区间[N,M]范围内的车道线识别置信度是比较可靠的；
应当指出的是：上述的置信度区间[N,M]是依据车道的变化而变化的，也即：车辆对于其所处车道的置信度区间和相邻车道的置信度区间的取值范围是不同的，例如，在一些实施例中，对于车辆所在的当前车道而言，N的取值范围为3-6米，M的取值范围为8-12米；对于当前车道的相邻车道，N的取值范围可以为9-11米，M的取值范围可以为13-17米。
在一些实施例中，引入最大车道线识别长度L，L代表先验的设定的最大车道线识别长度，即在任意时刻，车辆所识别的车道线长度不会大于L，L可以取50-100米范围内的数值。在一些实施例中，L可以取70米。
通过上述描述，应当理解：对于数据对准后的每一个时间戳，均具有一个和它相对应的车道线识别长度X和车道线识别置信度C。
还应当理解：对于数据对准后的每一个时间戳，车辆在空间上均有一个具体的、确定的车辆位置与它对应，而一旦确定了车辆所处坐标系，该车辆位置即可被用具体坐标的形式表示出来。
在一些实施例中，对上述的离线的、后验的数据按照时间戳逐个进行处理，以得到更新的车道线识别置信度。下面将以T时间戳为例，详细描述对于更新的车道线识别置信度获取的具体步骤：
参见图2，在T时间戳时刻，首先将车辆V的位置作为坐标系的原点，车辆行驶方向作为Y轴正向，与车辆行驶方向垂直的方向(即道路的宽度方向)作为X轴正向。需要指出的是：图2中为了示意清楚，将车辆V示意于X轴的正向部分而非零点，这仅是为了示例清楚的需要，而并不代表车辆V需要处于X轴的正向部分。还需要指出的是：图3中选取了车辆的几何中心点作为参考点，也可选车辆上的任意一点作为参考点，只要在后续的过程中均保持一致即可。
图2中还示出了在T时间戳时刻，车道线识别置信度区间N至M的示意，在一些实施例中，N可以取4米，而M可以取9米。
首先获取T时刻最大车道线识别长度L以及与最大车道线识别长度相对应的第一时间区间[S1,E1]，获取第一时间区间[S1,E1]的具体过程包括：
(1)在Y轴上获取自0点至正向L长度的范围，这段范围代表车辆在T时刻的最大车道线识别长度L，在一些实施例中，L可以取70米；
(2)在Y轴上获取自-N至L-M长度的范围，这样取值的原因在于：在T时刻，在Y轴0至N范围内的车道线相对于车辆而言实际上偏向于“侧方”而非“前方”，由于车载相机的视角/视场是有限的，车载相机对于偏向于侧方车道线的识别的可信度并不高，这也是为什么车道线识别置信度区间中N的取值不取零的原因。所以，为了获取T时刻车辆前方0至N范围内的车道线识别置信度，考虑在Y轴负向N处的车辆位置，如果车辆处于此位置，其识别的置信度区间在当前坐标系(T时刻车辆坐标系)下为[0,M-N]范围，在N和M分别取4和9(米)的情况下，上述[0,M-N]的范围完全涵盖了[0,N]的范围，因此可以采用与在Y轴负向N处的车辆位置所对应的时间戳的车道线识别置信度来“代替”车辆在T时刻的前方0 至N范围内的车道线识别置信度。类似地，对于远端(距离原点L米)而言，T时刻L-N至L范围内的置信度可以由距离远端L-M处的车辆位置所对应的时间戳的车道线识别置信度来代替；
(3)在上述(2)的过程中，由于在距离原点-N处以及L-M处不一定“恰好”就有与它们所对应的时间戳，因此考虑距离车辆-N±S MAX和L-M±S MAX的位置，其中S MAX代表两个相邻的时间戳之间车辆的最大位移，参见图2中A区域的放大图，由于车辆在相邻两帧之间的最大位移不超过S MAX，所以当将车辆位置(例如-N)向上或者向下移动了S MAX后(应当指出，向上或者向下移动都是可以的)，可以确保在移动后的车辆位置和处于-N位置的车辆位置之间必然有一个时间戳，在获取该时间戳后将其记录为S1，类似地，可以在远端获取时间戳E1；
应当理解的是：图2中的A区域放大图中，将车辆在X方向上进行了移动，这仅是为了在图中清晰地示出SMAX，而并不代表实际上车辆需要在X方向上进行这样的移动。
(4)基于时间戳S1和E1得到第一时间区间[S1,E1]，可以看出，与第一时间区间相对应的车辆空间位置区间是[-N±S,L-M±S]。
在确定了第一时间区间[S1,E1]后，即可获得在第一时间区间内的所有时间戳，记为序列P：{p 1，p 2，，...p n}。基于之前的记载应当理解：对于这个序列P中的每一个时间戳Pi(i＝1,2,3….n)，基于后验数据均可以获得一个与之对应的车道线识别长度Xi，车道线识别置信度C i；还可以确定在Pi时刻车辆所处的坐标位置；而Pi时刻的车道线识别置信度在车辆所处位置的行驶方向的前方[N,M]处是最可靠的。
对于每个时间戳Pi，判断该时刻的车道线识别长度Xi是否大于M，如果Xi大于M，则基于Xi获取第二时间区间[S2,E2]，基于和上述[S1,E1]中步骤(2)，(3)相似的原理，可以确定S2对应的位置为：车辆当前所处的位置的下方(Y轴负向/车辆行进方向的反方向)N±S处的位置，而E2对应的位置为，车辆当前所处位置的上方(Y轴正向)Xi-M±S的位置；与S2和E2对应的位置之间的空间范围即为与时间区间[S2,E2]相对应的空间距离。而当确定了第二时间区间[S2,E2]后，即可确认在这个时间戳时刻(Pi时刻)，车道线识别范围内所包括/对应的时间戳序列，记为序列Ki：{k i1，k i2，，...k im}。
遍历整个序列{p 1，p 2，，...p n}，对于其中每一个时间戳Pi，都获取与之对应的第二时间区间内所包括的时间戳序列Ki：{k i1，k i2，，...k im},i∈(1，n)。
对于相邻的或者相间隔的P序列中的元素(例如，但不限于P2，P4),与它们所分别对应的K2和K4序列中的元素可能会包括相重合的时间戳，也即是说，对于K2和K4序列中的某一个或者某几个时间戳而言，它们既可能出现在与P2对应的第二时间序列K2，也可能出现在与P4对应的第二时间序列K4。
遍历所有的与P序列的对应的第二时间序列(共有n个序列)，对于第二时间序列中的任意一个时间戳，判断它出现在多少个第二时间序列Ki中，将所有出现过的第二时间序列的对应P序列时间戳的车道线识别置信度求和，即得到该时间戳的更新的车道线识别置信度；例如，如果一个时间戳总共在与P1，P3，P4对应的第二时间序列中出现过三次，那么该时间戳的更新的车道线识别置信度即是P1，P3，P4的车道线识别置信度的求和。应当理解的是，这里的求和既可以是直接求和，也可以是加权求和。
在一些实施例中，在执行了上述的车道线识别置信度的求和后，还可以对求和后所获取 的更新的车道线识别置信度除以求和的数目以实现归一化。
应当指出的是，虽然上述过程中描述了两套时间序列P序列和Ki序列，但是Ki序列和P序列都是基于同一套时间戳数据而构建的，即在步骤S1中进行了数据对准后的时间戳数据。
参见图3来进一步理解上述过程，对于T时间戳，在确定了第一时间区间[S1,E1]后，即可确定第一时间区间内的所有时间戳{p 1，p 2，，...p n}，其在图4上示为Y轴上的多个圆点。图4示出了三个相邻的时间戳Pi，Pj，Pk并以空心圆点表示它们。基于每个时间戳，可以确定：
(1)该时刻车辆在Y方向位置，即该时刻圆点在Y轴上所处的位置；
(2)该时刻的车道线识别长度X，图4示出了分别与Pi，Pj，Pk对应的Xi，Xj，Xk；在图4的示例性示意中，Xi，Xj，Xk均大于M；图4中还示出了一个时间戳Pt，与Pt对应的车道线识别长度Xt小于M；
(3)基于Xi，Xj，Xk可以分别得到与它们对应的第二时间区间，与每个第二时间区间相对应的距离区间为图示的Li，Lj，Lk。例如，对于Pk，在获取了与之对应的Xk后，通过车辆当前所处的位置的下方(Y轴负向/车辆行进方向的反方向)N±S处的位置与车辆当前所处位置的上方(Y轴正向)X-M±S的位置之间的区间即可确定Lk，应当理解的是：图4中未直接示出N±S和Xk-M±S，而是示出了以当前车辆位置为参考点的[-N,Xk-M]区间作为示意，当确定了Li空间位置之后，即可依据时间和空间之间的对应关系确定与Xi对应的第二时间区间；
(4)和上述Xk相类似，对于Xi，Xj也可以分别得到Li和Lj。这样，对于这三个相邻的时间戳Pi，Pj，Pk即可分别得到与之相对应的第二时间区间，然后即可确定每个第二时间区间内所包括的时间戳，可以看出，对于Pi，Pj，Pk这三个时间戳而言，其都被三个区间Li，Lj，Lk所包括，因此在计算Pi，Pj，Pk这三个时间戳的任意一个的更新的车道线识别置信度时，Pi，Pj，Pk这三个时刻的车道线识别的置信度都必须被求和进来，应当理解的是，到此为止所讨论的仅仅是三个时间戳的情况，而如果还有其它的时间戳的第二时间区间也包括Pi，Pj，Pk，那么在计算它们的车道线识别置信度的时候，其它时间戳的车道线识别置信度也必须被求和进来。而对于Pe时间戳，其仅被区间Li包括，但是不被区间Lj和Lk所包括，所以在计算Pe时刻的车道线识别置信度时，不需要求和Pj和Pk时刻的车道线识别置信度。
(5)将图3中[S1,E1]区间内的所有时间戳均执行上述过程，并得到所有时间戳的更新的车道线线识别置信度，至此对于[S1,E1]时间区间的更新的车道线识别置信度计算结束；
(6)在获取了时间戳的更新的车道线识别置信度后，即可在空间上也确定车道线的不同分段的置信度。对于每一个时间戳而言，其更新的车道线识别置信度等同于在该时间戳时刻的车辆位置行进方向前方N至M米范围内的车道线识别置信度；应当理解的是：与不同时间戳相对应的的空间分段之间即有可能是有交叠的，也有可能是不交叠的；
(7)基于(5)、(6)应当理解，由于时间戳和车辆空间位置是相关联的，在每个时间戳时刻，都有与之对应的确定的车辆所处的空间位置(坐标值)。因此在(5)中，在时间上基于时间戳对于时间区间[S1,E1]进行分段并获取更新的与时间戳对 应的车道线识别置信度；而在(6)中，是在空间上基于和时间戳相对应的置信度区间对相对应的空间区间[-N±S,L-M±S]进行分段并获取更新的与空间分段相对应的车道线识别置信度，两者在实质上是等价的。
在上述过程中，对基于T时间戳时刻所确定的[S1,E1]区间内的每个时间戳而言，计算并获取了更新的车道线识别置信度。通过上述过程，可以提升每个时间戳的车道线识别置信度的可信程度/准确性。因为对于某一时间戳而言，现有技术中仅有一个车道线识别置信度的数值与之对应，而车道线识别置信度的准确性受到环境影响较大，例如，如果在某一时间戳时刻，车辆对向有车辆开着强光行驶，并且强光影响了车辆的摄像头对于前方环境的图像获取，则很有可能使得在该时间戳时刻的车道线识别置信度的结果会发生较大波动并因此而不准确；或者是车道线被其它的交通参与者(例如：车辆或者行人)所遮挡而无法被识别；在这些情况下，单个车道线识别置信度的值并不能“真正地”反映出识别结果。而本申请实施例中的技术方案，对于任一时间戳而言，通过对所有车道线识别范围包括它的时间戳(可以被记为/理解为和该任一时间戳相“对应”的时间戳集合)的车道线识别置信度的求和来确定该时间戳的更新的车道线识别置信度，使得由单一时间戳的识别波动所导致的车道线识别置信度的误差被最大化地消除/抑制。即对于一个时间戳(或者和该时间戳相对应的车道线识别置信度区间)而言，更新的车道线识别置信度在整体上体现了所有可能的车道线识别长度(或者和该车道线识别长度相对应的第二时间区间)包括该车道线识别置信度区间的时间戳的置信度；从而，实施例的技术方案可以抑制/消除由于识别波动所导致的车道线识别置信度的不准确性；提升了车道线识别置信度的可信程度/准确性。
在另外一些实施例中，可以对上述技术过程做适当的变化而不背离本申请的精神，例如，对于某一时间戳(或者和该时间戳相对应的车道线识别置信度区间)而言，可以使用两个、三个或者若干个车道线识别长度范围包括它的时间戳的车道线识别置信度的求和来获取更新的车道线识别置信度，而上述的两个、三个或若干个识别范围包括它的时间戳具体可以依据情况而选择，例如可以选择距离该时间戳最近邻的两个、三个或者若干个识别范围包括它的时间戳的车道线识别置信度进行求和来得到该时间戳的更新的车道线识别置信度，也可以在识别范围包括该时间戳的时间戳中选择两个、三个或者若干个识别范围最长的来确定该时间戳的更新的的车道线识别置信度；本领域技术人员可以依据实际需求而进行选择而不背离本申请的精神。
图4示例性地总结了上述技术流程400，具体包括：
401：获取T时刻最大车道线识别长度以及与最大识别长度相对应的第一时间区间；
402：遍历第一时间区间内的每个时间戳，获取与每个时间戳相对应的车道识别长度；
403：对每个时间戳，判断与它相对应的车道线识别长度是否大于M；如果判断结果为否，则继续返回402获取下一个时间戳以及与它相对应的车道线识别长度；如果判断结果为是，则进入404；
404：获取与车道线识别长度相对应的第二时间区间；
405：对于第一时间区间内的每一个时间戳，基于多个车道线识别长度(或者和车道线识别长度相对应的第二时间区间)包括它的时间戳的车道线识别置信度的求和来获取更新的车道线识别置信度。
在另外一些实施例中，可以将上述技术过程延伸到实时场景，即车辆正在行驶的场景，在此种情况下，可以选取例如车道线识别范围包括当前时间戳的多个前序的(时间上在当前 时间戳之前的)时间戳(可以是两个、三个或更多个)的车道线识别置信度求和来获取当前时间戳的更新的车道线识别置信度，从而可以实时地提升车道线识别置信度的可信程度/准确性。
S3：车道线真值构建
在完成了S2后，即可进入S3，即车道线真值构建步骤，该步骤的目的是基于S2所获得的更新的车道线识别置信度来构建车道线真值。
仍以T时间戳为例，在S2中获取了T时间戳的第一时间区间[S1,E1]以及第一时间区间内的所有时间戳{p 1，p 2，，...p n}的更新的车道线识别置信度。
参见图5，其示例性地示出了车道线真值构建的流程500，包括：
501：首获取第一时间区间及其中的时间戳，遍历第一时间区间内的时间戳；
502：对于每一时间戳，判断与该时间戳对应的车道线识别置信度是否大于设定阈值。如果大于设定阈值，则进入下一步503；否则返回上一步获取下一个时间戳；在一些实施例中，可以将阈值设为0.6，应当理解的是，技术人员可以依据实际需求而设定阈值的数值而不背离本申请的精神；
503：判断与时间戳对应的车道线识别长度是否大于M，如果车道线识别长度小于M，将进入离线车道线跟踪流程508(图5中以虚线框示出)；如果车道线识别长度大于M，则进入步骤504；
如上所述：如果在503中判断某一时间戳(例如Pi)的车道线识别长度小于设定阈值M，则转入图5右侧所示的离线车道线跟踪流程508，流程508包括：
5081：首先获取与Pi所对应的第二时间区间，然后依据时间顺序遍历第二时间区间中的不同于当前时间戳Pi的其它时间戳；
5082：对于所遍历的其它时间戳中的每一个而言，将与其对应的车道线识别长度乘以设定系数；这里之所以设置设定系数的意义在于，该步骤是以其它时间戳的车道线识别长度(结果)来代替Pi时刻的车道线识别长度，由于将其它时间戳的车道线识别长度转换到Pi时间戳后，会有置信度的损失，所以这里使用设定系数来表征这一损失，设定系数可以选取0.4至0.6范围内的值；
5083：判断步骤5082所获得的值是否大于M，如果大于M，则进入步骤504，反之则返回到上一步继续遍历下一个时间戳；
504：获取该时间戳时刻距离车辆N至M米范围内的车道线点集，并依据惯性导航装置的位置和航向角将该时间戳时刻的车道线点集映射转换到T时刻的车辆坐标系；
505：对所获得的车道线点集进行聚类和分组，可以使用习知的聚类方法来对车道线点集进行聚类分组，例如可以使用DBSCAB(Denisty-Based Spatial Clustering of Application with Noise)方法，应当理解的是，其它任何合适的聚类方法也可以被使用；
506：当分组完成后，基于分组后的车道线点集组之间的纵向聚类和车道线存在性判断是否进行共线连接；具体包括：获取分组后的车道线点集结果，如果分组之间的车道线的纵向距离/间隔大于设定阈值Q(Q可以取20至30米中的任意值)，则进行共线连接判断，共线连接判断包括两个步骤，第一步：将间隔两侧的车道线向量化，并判断两个向量之间的夹角或者横向间距，如果夹角超过设定阈值(例如30度)，或者横向间距大于设定阈值(例如3米)，则判断无需进行共线连接；而如果判断两个向量之间的夹角和横向间距均小于设定阈值，则进入第二步，即考虑在间隔范围内的时间戳的车道线识别置信度，如果间隔范围内的 多数时间戳的车道线识别置信度小于设定阈值(例如上述的0.6)，则仍然不进行共线连接，而当间隔范围内的多数时间戳的车道线识别置信度大于设定阈值时，则进行共线连接；
在步骤507，对506所获得的车道线进行多项式拟合，获取T时间戳时刻的车道线真值。
对于其他的时间戳，可以采用和上述相一致的方案获取车道线真值。
S4：车道线识别异常事件提取
在S3获取车道线真值后，即可用于校验车道线识别结果，比较对象包括：经由S3获得的车道线真值以及车辆所获得的车道线识别结果，车道线识别结果可以包括由架构在云上的车道线识别算法基于车辆所获得的图像信息而得到的。可以依据预设的指标来进行评价，预设的指标可以包括：(1)车道线识别结果过短；(2)车道线识别结果横向误差过大；(3)车道线识别结果朝向误差过大；(4)车道线识别结果漏检。
在一些实施例中，可以采用多项式对车道线进行拟合，具体而言，可以使用三次多项式对车道线识别结果和车道线真值分别进行拟合，拟合后的车道线识别结果为：y＝c(x)＝c 0+c 1c+c 2x 2+c 3c 3，拟合后的车道线真值为＝m(x)＝m 0+m 1x+mx 2+m 3x 3。
对于每一帧T而言，定义车道线真值长度为Ym，车道线识别长度为Yc(米)，并定义截断点为x 1，x 2，，...x n，截断点指在车道线识别结果轨迹上预设的点，例如距离原点10米、20米、30米处的点，本领域技术人员可以依据实际情况而具体选择合适的截断点。
在一些实施例中，基于以上，预设的指标的评价过程和含义如下：
(1)如果Yc/Ym小于设定阈值，则认为车道线识别结果过短，在一些实施例中，这里的设定阈值可以取0.5，应当理解的是，本领域技术人员可以依据实际情况来选择合适的阈值而不背离本申请的精神；
(2)计算多项式在各个截断点x 1，x 2，，...x n处的y值，若在任意一个截断点处y方向误差(即|c(x i)-m(x i)|)大于设定阈值，则认为车道线识别结果横向误差过大，在一些实施例中，这里的设定阈值看可以取1米，应当理解的是，本领域技术人员可以依据实际情况来选择合适的阈值而不背离本申请的精神；
(3)计算多项式在各个截断点处的y值，对于每一个截断点，构建以车道线起点为起点，截断点为终点的向量(即
Figure PCTCN2020085470-appb-000001
比较两个向量的夹角，若存在任意一组向量之间的夹角大于设定阈值，则车道线识别结果朝向误差过大，在一些实施例中，这里的设定阈值看可以取30度，应当理解的是，本领域技术人员可以依据实际情况来选择合适的阈值而不背离本申请的精神；
(4)车道线识别结果漏检：存在车道线真值Ym而车道线识别结果Ym没有输出的情况。
在以上几种情况中，通过比较本申请实施例通过后验数据获取的车道线真值和车道线识别结果，可以确定出车道线识别异常的情况，而这些识别异常的情况通过现有的车道线识别算法自身无法得出。进一步地，这些识别异常的情况可以作为训练材料被反馈到车道线识别算法(往往采用神经深度学习算法)，从而完善车道线识别算法的训练材料，提升车道线识别算法的识别精度。
S5：将识别出的异常事件上传至云端
在一些实施例中，车道线识别算法可以设置在服务端，车道线识别异常的数据可以传输给服务端的车道线识别算法用以提升算法的识别精度。在一些实施例中，服务端可以是云端，在一些实施例中，服务端也是可以服务器。云端的车道线识别算法在训练后也可以通过通信 网络为车端提供更精确/可靠的车道线识别能力。如此循环，可以形成良性的数据-训练材料-算法性能提升的循环。
S6:结束
一些实施例还提供了一种车道线识别装置，参见图6，车道线识别装置包括设置在车端的车道线识别模块、惯性导航模块、计算装置，计算装置和车道线识别模块以及惯性导航模块通信连接并被配置为可以执行上述S0-S6的各种实施例的方案。
一些实施例还提供了一种车道线识别系统，继续参见图6，系统包括设置车端的车道线识别装置和设置在服务端的车道线识别算法模块；车道线识别装置包括车道线识别模块、惯性导航模块、计算装置；车道线识别装置和车道线识别算法模块通信连接，所述车道线识别装置被配置为将车道线识别异常事件发送给所述车道线识别算法模块，所述车道线识别算法模块被配置为可使用车道线识别异常事件对所述车道线识别算法进行训练，以优化车道线识别算法的识别精度。在一些实施例中，服务端可以是云端，在一些实施例中，服务端也是可以服务器。
一些实施例还提供一种计算机可读存储介质，参见图7，计算机可读存储介质71包括指令集73，指令集73可以被处理器执行以实现上述S0-S6的各种实施方案。
一些实施例还提供一种车道线识别系统，参见图7，车道线识别系统包括处理器72和计算机可读存储介质71，计算机可读存储介质存储有指令集73，指令集73可以被处理器执行以实现上述S0-S6的各种实施方案。
本申请各种实施例提供了一种车辆可行驶区域检测方法、系统以及采用该系统的自动驾驶车辆。在本申请技术方案中，通过使用多种传感器，并融合基于多种传感器所获得的障碍物概率信息以获得概率栅格地图，从而可以更加准确地识别车辆周围的可行驶区域；另一方面，本申请利用了障碍物视差信息、障碍物贴图等方式，使得本申请的技术方案可以具有较高的可泛化性，广泛适用于多种场景，并且不依赖于训练数据，因此本申请技术方案具有鲁棒性。综上：本申请技术方案可以广泛地适用于不同等级的自动驾驶方案、系统、车辆。
本申请的说明书和权利要求书及上述附图中的术语“第一”、“第二”、“第三”、“第四”等(如果存在)是用于区别类似的对象，而不必用于描述特定的顺序或先后次序。应该理解这样使用的数据在适当情况下可以互换，以便这里描述的实施例能够以除了在这里图示或描述的内容以外的顺序实施。此外，术语“包括”和“具有”以及他们的任何变形，意图在于覆盖不排他的包含，例如，包含了一系列步骤或单元的过程、方法、系统、产品或设备不必限于清楚地列出的那些步骤或单元，而是可包括没有清楚地列出的或对于这些过程、方法、产品或设备固有的其它步骤或单元。
所属领域的技术人员可以清楚地了解到，为描述的方便和简洁，上述描述的系统，装置和单元的具体工作过程，可以参考前述方法实施例中的对应过程，在此不再赘述。
在本申请所提供的几个实施例中，应该理解到，所揭露的系统，装置和方法，可以通过其它的方式实现。例如，以上所描述的装置实施例仅仅是示意性的，例如，单元的划分，仅仅为一种逻辑业务划分，实际实现时可以有另外的划分方式，例如多个单元或组件可以结合或者可以集成到另一个系统，或一些特征可以忽略，或不执行。另一点，所显示或讨论的相互之间的耦合或直接耦合或通信连接可以是通过一些接口，装置或单元的间接耦合或通信连接，可以是电性，机械或其它的形式。
作为分离部件说明的单元可以是或者也可以不是物理上分开的，作为单元显示的部件可 以是或者也可以不是物理单元，即可以位于一个地方，或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部单元来实现本实施例方案的目的。
另外，在本申请各个实施例中的各业务单元可以集成在一个处理单元中，也可以是各个单元单独物理存在，也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现，也可以采用软件业务单元的形式实现。
集成的单元如果以软件业务单元的形式实现并作为独立的产品销售或使用时，可以存储在一个计算机可读取存储介质中。基于这样的理解，本申请的技术方案本质上或者说对现有技术做出贡献的部分或者该技术方案的全部或部分可以以软件产品的形式体现出来，该计算机软件产品存储在一个存储介质中，包括若干指令用以使得一台计算机设备(可以是个人计算机，服务器，或者网络设备等)执行本申请各个实施例方法的全部或部分步骤。而前述的存储介质包括：U盘、移动硬盘、只读存储器(ROM，Read-Only Memory)、随机存取存储器(RAM，Random Access Memory)、磁碟或者光盘等各种可以存储程序代码的介质。
本领域技术人员应该可以意识到，在上述一个或多个示例中，本申请所描述的业务可以用硬件、软件、固件或它们的任意组合来实现。当使用软件实现时，可以将这些业务存储在计算机可读介质中或者作为计算机可读介质上的一个或多个指令或代码进行传输。计算机可读介质包括计算机存储介质和通信介质，其中通信介质包括便于从一个地方向另一个地方传送计算机程序的任何介质。存储介质可以是通用或专用计算机能够存取的任何可用介质。
以上的具体实施方式，对本申请的目的、技术方案和有益效果进行了进一步详细说明，所应理解的是，以上仅为本申请的具体实施方式而已。
以上实施例仅用以说明本申请的技术方案，而非对其限制；尽管参照前述实施例对本申请进行了详细的说明，本领域的普通技术人员应当理解：其依然可以对前述各实施例所记载的技术方案进行修改，或者对其中部分技术特征进行等同替换；而这些修改或者替换，并不使相应技术方案的本质脱离本申请各实施例技术方案的范围。

Claims (16)

  1. 一种车道线识别置异常事件确定方法，包括：
    至少依据后验的车道线识别结果确定更新的车道线识别置信度；
    至少依据所述更新的车道线识别置信度构建车道线真值；
    依据所述车道线识别结果和所述车道线真值确定车道线识别异常事件；
    其中，所述后验的车道线识别结果包括当车辆行驶完成后所获取的车道线识别置信度。
  2. 根据权利要求1所述的车道线识别异常事件确定方法，其中：
    所述至少依据后验的车道线识别结果确定更新的车道线识别置信度包括：将所述后验的车道线识别结果和后验的惯性导航数据基于时间戳进行对准；其中，所述后验的惯性导航数据包括当车辆行驶完成后所获取的惯性导航数据。
  3. 根据权利要求2所述的车道线识别异常事件确定方法，其中：
    所述时间戳具有与其对应的车道线识别结果，所述车道线识别结果包括：车辆位置、车道线识别置信度和车道线识别长度。
  4. 根据权利要求3所述的车道线识别异常事件确定方法，其中：
    所述至少依据后验的车道线识别结果确定更新的车道线识别置信度包括：对于任一时间戳，至少基于与所述任一时间戳对应的所述后验的车道线识别结果中的两个车道线识别置信度确定与所述任一时间戳对应的更新的车道线识别置信度。
  5. 根据权利要求4所述的车道线识别异常事件确定方法，其中：
    所述基于与所述任一时间戳对应的所述后验的车道线识别结果中的两个车道线识别置信度确定与所述任一时间戳对应的更新的车道线识别置信度包括：
    对于任一时间戳，确定与所述任一时间戳对应的时间戳集合，所述时间戳集合中包括一个或多个时间戳，所述一个或多个时间戳的车道线识别长度均包括与所述任一时间戳对应的车辆位置，获取所述时间戳集合中每个时间戳对应的车道线识别置信度并构成车道线识别置信度集合，从所述车道线识别置信度集合中获取至少两个进行求和以得到与所述任一时间戳对应的更新的车道线识别置信度。
  6. 根据权利要求5所述的车道线识别异常时间确定方法，其中：
    所述从所述车道线识别置信度中获取至少两个进行求和以得到与所述任一时间戳对应的更新的车道线识别置信度包括：
    对所述车道线识别置信度集合中的所有车道线识别置信度进行求和以得到与所述任一时间戳对应的更新的车道线识别置信度。
  7. 根据权利要求6所述的车道线识别异常事件确定方法，其中：
    所述求和包括直接求和与加权求和中的至少一个。
  8. 根据权利要求3-7中任一所述的车道线识别异常事件确定方法，其中：
    所述至少依据所述更新的车道线识别置信度构建车道线真值包括：
    判断更新的车道线识别置信度是否大于第一阈值，以及车道线识别长度是否大于第二阈值，获取第一时间戳集合，所述第一时间戳集合中的每个时间戳对应的车道线识别置信度大于第一阈值，并且所述第一时间戳集合中的每个时间戳对应的车道线识别长度大于第一阈值；
    对所述第一时间戳集合中的每个时间戳，获取车辆坐标系下车辆近端N至M米的第一车道线点集。
  9. 根据权利要求8所述的车道识别异常事件确定方法，还包括：
    获取更新的车道线识别置信度大于第一阈值且车道线识别长度小于第二阈值的第二时间戳集合；
    对于第二时间戳集合中的每个时间戳，以符合设定条件的其它时间戳来代替它，所述其他时间戳对应的车道线识别长度大于第一阈值；
    对代替后的第二时间戳集合中的每个时间戳，均获取车辆坐标系下车辆近端N至M米的第二车道线点集。
  10. 根据权利要求9所述的车道线识别异常事件确定方法，还包括：
    对所述第一车道线点集和第二车道线点集进行聚类分组，如果分组后车道线点集组之间的纵向距离大于距离阈值，则基于车道线存在性结果的判断决定是否进行共线连接。
  11. 根据权利要求10所述的车道线识别异常事件确定方法，还包括：
    将共线连接后的车道线点集进行多项式拟合，获得车道线真值。
  12. 根据权利要求8-11任一所述的车道线识别异常事件确定方法，其中：
    所述依据所述车道线识别结果和所述车道线真值确定车道线识别异常事件包括：
    比较所述车道线真值和所述车道线识别结果以确定异常事件，所述异常事件包括：车道线识别结果过短、车道线识别结果横向误差过大、车道线识别结果朝向误差过大，车道线识别结果漏检。
  13. 根据权利要求12所述的方法，还包括：
    将所述异常事件上传服务端，所述异常事件用于训练所述服务端的车道线识别算法。
  14. 一种车道线识别装置，包括：
    设置于车端的车道线识别模块；
    设置于车端的惯性导航模块；
    设置于车端并与所述车道线识别模块和惯性导航模块通信连接的计算装置；
    其中，所述计算装置被配置为可执行权利要求1-13任一所述的方法。
  15. 一种车道线识别系统，包括：
    如权利要求14所述的车道线识别装置；
    设置在服务端的车道线识别算法模块，所述车道线算法识别模块包括车道线识别算法；
    所述车道线识别装置和所述车道线识别算法模块通信连接，所述车道线识别装置被配置为将所述异常事件发送给所述车道线识别算法模块，所述车道线识别算法模块被配置为使用所述异常事件对所述车道线识别算法进行训练。
  16. 一种计算机可读存储介质，包括指令集，所述指令集可以被处理器执行以实现如权利要求1-13任一所述的方法。
PCT/CN2020/085470 2020-04-18 2020-04-18 车道线识别异常事件确定方法、车道线识别装置及系统 WO2021208110A1 (zh)

Priority Applications (4)

Application Number Priority Date Filing Date Title
PCT/CN2020/085470 WO2021208110A1 (zh) 2020-04-18 2020-04-18 车道线识别异常事件确定方法、车道线识别装置及系统
EP20931648.8A EP4130668A4 (en) 2020-04-18 2020-04-18 METHOD FOR DETERMINING AN ABNORMAL LANE LINE RECOGNITION EVENT, AND LANE LINE RECOGNITION APPARATUS AND SYSTEM
CN202080004432.3A CN112639765B (zh) 2020-04-18 2020-04-18 车道线识别异常事件确定方法、车道线识别装置及系统
US17/967,376 US20230034574A1 (en) 2020-04-18 2022-10-17 Method for determining lane line recognition abnormal event, and lane line recognition apparatus and system

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
PCT/CN2020/085470 WO2021208110A1 (zh) 2020-04-18 2020-04-18 车道线识别异常事件确定方法、车道线识别装置及系统

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US17/967,376 Continuation US20230034574A1 (en) 2020-04-18 2022-10-17 Method for determining lane line recognition abnormal event, and lane line recognition apparatus and system

Publications (1)

Publication Number Publication Date
WO2021208110A1 true WO2021208110A1 (zh) 2021-10-21

Family

ID=75291152

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/085470 WO2021208110A1 (zh) 2020-04-18 2020-04-18 车道线识别异常事件确定方法、车道线识别装置及系统

Country Status (4)

Country Link
US (1) US20230034574A1 (zh)
EP (1) EP4130668A4 (zh)
CN (1) CN112639765B (zh)
WO (1) WO2021208110A1 (zh)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115049997A (zh) * 2022-06-07 2022-09-13 北京百度网讯科技有限公司 边缘车道线的生成方法、装置、电子设备及存储介质
CN115071733A (zh) * 2022-07-21 2022-09-20 成都工业职业技术学院 一种基于计算机的辅助驾驶方法及装置

Families Citing this family (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
KR20220013203A (ko) * 2020-07-24 2022-02-04 현대모비스 주식회사 차량의 차선 유지 보조 시스템 및 이를 이용한 차선 유지 방법
CN113758501A (zh) * 2021-09-08 2021-12-07 广州小鹏自动驾驶科技有限公司 检测地图中的异常车道线的方法和可读存储介质
CN117490728B (zh) * 2023-12-28 2024-04-02 合众新能源汽车股份有限公司 车道线定位故障诊断方法和系统

Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102385698A (zh) * 2011-10-26 2012-03-21 潘承志 基于置信度传播的图像处理方法
CN105046235A (zh) * 2015-08-03 2015-11-11 百度在线网络技术（北京）有限公司 车道线的识别建模方法和装置、识别方法和装置
CN105667518A (zh) * 2016-02-25 2016-06-15 福州华鹰重工机械有限公司 车道检测的方法及装置
JP2019078562A (ja) * 2017-10-20 2019-05-23 トヨタ自動車株式会社 自車位置推定装置
JP2019132762A (ja) * 2018-02-01 2019-08-08 トヨタ自動車株式会社 自車位置推定装置
CN110160540A (zh) * 2019-06-12 2019-08-23 禾多科技（北京）有限公司 基于高精度地图的车道线数据融合方法

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US10215583B2 (en) * 2013-03-15 2019-02-26 Honda Motor Co., Ltd. Multi-level navigation monitoring and control
US9145116B2 (en) * 2013-12-04 2015-09-29 Mobileye Vision Technologies Ltd. Systems and methods for implementing a multi-segment braking profile for a vehicle
CN104792792B (zh) * 2015-04-27 2017-10-20 武汉武大卓越科技有限责任公司 一种逐步求精的路面裂缝检测方法
CN105701449B (zh) * 2015-12-31 2019-04-23 百度在线网络技术（北京）有限公司 路面上的车道线的检测方法和装置
US10606259B2 (en) * 2017-07-07 2020-03-31 Zoox, Inc. Interactions between vehicle and teleoperations system
WO2019068042A1 (en) * 2017-09-29 2019-04-04 Netradyne Inc. MULTIPLE EXPOSURE EVENT DETERMINATION
US10997431B2 (en) * 2018-06-05 2021-05-04 GM Global Technology Operations LLC Controlling a vehicle using an alternate anchor line
CN109724615B (zh) * 2019-02-28 2021-06-29 北京经纬恒润科技股份有限公司 一种车道线识别结果的校验方法及系统

Patent Citations (6)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN102385698A (zh) * 2011-10-26 2012-03-21 潘承志 基于置信度传播的图像处理方法
CN105046235A (zh) * 2015-08-03 2015-11-11 百度在线网络技术（北京）有限公司 车道线的识别建模方法和装置、识别方法和装置
CN105667518A (zh) * 2016-02-25 2016-06-15 福州华鹰重工机械有限公司 车道检测的方法及装置
JP2019078562A (ja) * 2017-10-20 2019-05-23 トヨタ自動車株式会社 自車位置推定装置
JP2019132762A (ja) * 2018-02-01 2019-08-08 トヨタ自動車株式会社 自車位置推定装置
CN110160540A (zh) * 2019-06-12 2019-08-23 禾多科技（北京）有限公司 基于高精度地图的车道线数据融合方法

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
See also references of EP4130668A4 *

Cited By (3)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115049997A (zh) * 2022-06-07 2022-09-13 北京百度网讯科技有限公司 边缘车道线的生成方法、装置、电子设备及存储介质
CN115071733A (zh) * 2022-07-21 2022-09-20 成都工业职业技术学院 一种基于计算机的辅助驾驶方法及装置
CN115071733B (zh) * 2022-07-21 2022-10-25 成都工业职业技术学院 一种基于计算机的辅助驾驶方法及装置

Also Published As

Publication number Publication date
US20230034574A1 (en) 2023-02-02
EP4130668A4 (en) 2023-05-31
CN112639765A (zh) 2021-04-09
EP4130668A1 (en) 2023-02-08
CN112639765B (zh) 2022-02-11

Similar Documents

Publication Publication Date Title
WO2021208110A1 (zh) 车道线识别异常事件确定方法、车道线识别装置及系统
CN112417967B (zh) 障碍物检测方法、装置、计算机设备和存储介质
US20220026232A1 (en) System and method for precision localization and mapping
WO2023124383A1 (zh) 一种车辆速度检测、撞车预警方法及电子设备
CN109272756B (zh) 一种信控交叉口排队长度估计方法
CN109084786B (zh) 一种地图数据的处理方法
WO2021249071A1 (zh) 一种车道线的检测方法及相关设备
US20220410939A1 (en) Collision detection method, electronic device, and medium
US20210229280A1 (en) Positioning method and device, path determination method and device, robot and storage medium
CN112380312B (zh) 基于栅格检测的激光地图更新方法、终端及计算机设备
RU2764708C1 (ru) Способы и системы для обработки данных лидарных датчиков
CN114170797A (zh) 交通限制路口的识别方法、装置、设备、介质及产品
EP4020428A1 (en) Method and apparatus for recognizing lane, and computing device
CN113111707A (zh) 一种基于卷积神经网络的前车检测与测距方法
CN110081890B (zh) 一种结合深度网络的动态k最近邻地图匹配方法
CN112530159A (zh) 一种自标定式多车道级交通流量检测方法及电子设备
CN112100565A (zh) 一种道路曲率确定方法、装置、设备及存储介质
US20230154203A1 (en) Path planning method and system using the same
WO2022193448A1 (zh) 一种定位方法、装置、电子设备及存储介质
CN112130137A (zh) 确定车道级轨迹的方法、装置及存储介质
CN115662167B (zh) 自动驾驶地图构建方法、自动驾驶方法及相关装置
US20170010103A1 (en) Map creation system and map creating method
WO2023097873A1 (zh) 用于车辆定位检查的方法、装置、存储介质及设备
CN116878487B (zh) 一种用于建立自动驾驶地图的方法、装置、车辆和服务器
CN117392166B (zh) 一种基于地平面拟合的三阶段点云地面分割方法

Legal Events

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

Ref document number: 20931648

Country of ref document: EP

Kind code of ref document: A1

ENP Entry into the national phase

Ref document number: 2020931648

Country of ref document: EP

Effective date: 20221025

NENP Non-entry into the national phase

Ref country code: DE