Positioning accuracy evaluation method, device and vehicle

Abstract

Embodiments of the present application disclose a positioning accuracy evaluation method and device and a vehicle. The method comprises: acquiring laser mapping data, the laser mapping data comprising a laser point cloud map; acquiring vision mapping data, the vision mapping data comprising a vision point cloud map; acquiring a first error parameter between the laser point cloud map and the vision point cloud map, and acquiring a second error parameter between the laser point cloud map and the vision point cloud map, the point cloud scale corresponding to the first error parameter being greater than the point cloud scale corresponding to the second error parameter, the point cloud scale representing the size of an area occupied by an object formed by a point cloud in the point cloud map; and evaluating the positioning accuracy of a preset vision mapping algorithm according to the first error parameter and the second error parameter. The method can directly evaluate the output result of the vision mapping algorithm according to the map data used for positioning, thereby improving the effectiveness of the positioning accuracy evaluation.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent driving, and more particularly, to a positioning accuracy evaluation method and device and a vehicle. BACKGROUND

[0002] With the gradual development of machine vision and other technologies, SLAM (Simultaneous Localization and Mapping) technology is widely used in various intelligent driving scenarios. In the intelligent driving scenario, by introducing the SLAM technology, scene positioning or environment perception and other function supports can be provided for the intelligent vehicle to perform subsequent processes such as path planning and driving control in the scene, especially to provide function support for intelligent parking tasks. However, the current visual mapping algorithm mostly relies on high-precision sensors to obtain data, which is high in cost and difficult to directly and effectively evaluate the effect of visual mapping. SUMMARY

[0003] In view of the above problems, the present application provides a positioning accuracy evaluation method, device and vehicle to improve the above problems.

[0004] In a first aspect, the present application provides a positioning accuracy evaluation method, which comprises: obtaining laser mapping data, the laser mapping data comprising a laser point cloud map; obtaining visual mapping data, the visual mapping data comprising a visual point cloud map; obtaining a first error parameter between the laser point cloud map and the visual point cloud map, and obtaining a second error parameter between the laser point cloud map and the visual point cloud map, the point cloud scale corresponding to the first error parameter being greater than the point cloud scale corresponding to the second error parameter, the point cloud scale representing the size of the region occupied by the object composed of the point cloud in the point cloud map; and evaluating the positioning accuracy of a preset visual mapping algorithm according to the first error parameter and the second error parameter.

[0005] In a second aspect, the present application provides a positioning accuracy evaluation device, the device comprising: a laser mapping result acquisition module, configured to acquire laser mapping data, the laser mapping data comprising a laser point cloud map; a vision mapping result acquisition module, configured to acquire vision mapping data, the vision mapping data comprising a vision point cloud map; an error parameter acquisition module, configured to acquire a first error parameter between the laser point cloud map and the vision point cloud map, and to acquire a second error parameter between the laser point cloud map and the vision point cloud map, the first error parameter corresponding to a point cloud scale larger than the second error parameter corresponding to a point cloud scale, the point cloud scale representing the size of the region occupied by the object composed of the point cloud in the point cloud map; and a positioning accuracy evaluation module, configured to evaluate the positioning accuracy of a preset vision mapping algorithm according to the first error parameter and the second error parameter.

[0006] In a third aspect, the present application provides a vehicle comprising one or more processors and a memory; one or more programs stored in the memory and configured to be executed by the one or more processors, the one or more programs configured to perform the above method.

[0007] In a fourth aspect, the present application provides a computer readable storage medium, the computer readable storage medium storing a program code, wherein the program code performs the above method when running.

[0008] In a fifth aspect, the present application provides a computer program product comprising computer programs / instructions, which, when executed by a processor, implement the steps of the above method.

[0009] The positioning accuracy evaluation method, device, vehicle and storage medium provided by the present application, by acquiring laser mapping data, the laser mapping data comprising a laser point cloud map; acquiring vision mapping data, the vision mapping data comprising a vision point cloud map; acquiring a first error parameter between the laser point cloud map and the vision point cloud map, and acquiring a second error parameter between the laser point cloud map and the vision point cloud map, the first error parameter corresponding to a point cloud scale larger than the second error parameter corresponding to a point cloud scale, the point cloud scale representing the size of the region occupied by the object composed of the point cloud in the point cloud map; and evaluating the positioning accuracy of a preset vision mapping algorithm according to the first error parameter and the second error parameter. Thus, by acquiring the errors in two different scale ranges between the laser point cloud map and the vision point cloud map, and then evaluating the positioning accuracy of the preset vision mapping algorithm based on the sum of the two errors, the output result of the vision mapping algorithm can be directly evaluated according to the map data used for positioning, and the effectiveness of the positioning accuracy evaluation is improved. At the same time, since no expensive high-precision sensor is introduced to acquire data, the evaluation cost of the positioning accuracy can be reduced. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of this application, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 A flowchart of a positioning accuracy evaluation method proposed in an embodiment of this application is shown.

[0012] Figure 2 An example diagram of the installation method of the lidar provided in the embodiments of this application is shown.

[0013] Figure 3 An example diagram of the LiDAR-image 3D checkerboard calibration board provided in an embodiment of this application is shown.

[0014] Figure 4 It shows Figure 1 A flowchart of a method for step S130.

[0015] Figure 5 A schematic flowchart illustrating the acquisition of the first error parameter between the laser point cloud map and the visual point cloud map provided in an embodiment of this application is shown.

[0016] Figure 6 An example diagram is shown illustrating planar detection using laser map data and visual map data, as provided in an embodiment of this application.

[0017] Figure 7 An example diagram is shown illustrating the extraction of structured regions from laser map data and visual map data according to an embodiment of this application.

[0018] Figure 8 It shows Figure 1 Another method flowchart for step S130 in the process.

[0019] Figure 9 An example diagram is shown, illustrating the acquisition of a first 3D bounding box corresponding to a first target point cloud and a second 3D bounding box corresponding to a second target point cloud, provided by an embodiment of this application.

[0020] Figure 10 A schematic flowchart illustrating the acquisition of a second error parameter between a laser point cloud map and a visual point cloud map, provided in an embodiment of this application, is shown.

[0021] Figure 11Another example diagram of obtaining a first three-dimensional bounding box corresponding to a first target point cloud and obtaining a second three-dimensional bounding box corresponding to a second target point cloud is shown.

[0022] Figure 12 A flow chart of a positioning accuracy evaluation method is shown.

[0023] Figure 13 An example flow chart of obtaining adjacent frames with time intervals greater than a specified time interval and distance intervals greater than a specified distance interval in a laser mapping trajectory is shown.

[0024] Figure 14 A curve diagram of a plurality of absolute trajectory errors calculated by the embodiments of the present application is shown.

[0025] Figure 15 A curve diagram of a plurality of relative trajectory errors calculated by the embodiments of the present application is shown.

[0026] Figure 16 A structural block diagram of a positioning accuracy evaluation device is shown.

[0027] Figure 17 A structural block diagram of a vehicle is shown. DETAILED DESCRIPTION

[0028] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. All other embodiments obtained by those skilled in the art based on the embodiments of the present application without creative labor fall within the scope of protection of the present application.

[0029] In the embodiments of the present application, the inventors propose a positioning accuracy evaluation method, device and vehicle. The laser mapping data is obtained, and the laser mapping data includes a laser point cloud map. The vision mapping data is obtained, and the vision mapping data includes a vision point cloud map. The first error parameter between the laser point cloud map and the vision point cloud map is obtained, and the second error parameter between the laser point cloud map and the vision point cloud map is obtained. The point cloud scale corresponding to the first error parameter is larger than the point cloud scale corresponding to the second error parameter. The point cloud scale represents the size of the region occupied by the object composed of the point cloud in the point cloud map. The positioning accuracy of the preset vision mapping algorithm is evaluated according to the first error parameter and the second error parameter. Thus, by obtaining the errors in two different scale ranges between the laser point cloud map and the vision point cloud map, and then evaluating the positioning accuracy of the preset vision mapping algorithm based on the sum of the two errors, the output result of the vision mapping algorithm can be directly evaluated according to the map data used for positioning, and the effectiveness of the positioning accuracy evaluation is improved. At the same time, since the expensive high-precision sensor is not introduced to obtain data, the evaluation cost of the positioning accuracy can be reduced.

[0030] Please refer to Figure 1 A positioning accuracy evaluation method is provided in an embodiment of the present application. The positioning accuracy evaluation method can be applied to the intelligent parking scene of an indoor parking lot, and can be specifically used for evaluating the positioning accuracy of an intelligent parking task. The method comprises the following steps.

[0031] S110: Obtain laser mapping data, wherein the laser mapping data includes a laser point cloud map.

[0032] In the embodiments of the present application, the laser mapping data represents the data obtained after laser mapping by laser radar data. In an implementation manner, the laser mapping data can include a laser point cloud map (i.e., a map composed of point cloud data obtained by laser radar scanning).

[0033] As an implementation manner, the laser mapping data can be obtained by a reference sensor. Optionally, the reference sensor can include a laser radar, which can be a low-cost laser radar independent of external signals. For example, the laser radar can be a 128-line laser radar. It should be noted that the type of laser radar herein is only an example and does not constitute a limitation on the type of laser radar.

[0034] As a specific implementation manner, before obtaining the laser mapping data, the laser radar (i.e., the reference sensor) can be first mounted on the vehicle in an external manner through a fixed support, as shown in Figure 2 Figure 2 ​The laser radar is mounted on the roof of the vehicle, and is connected to a network router port in the vehicle, so that the laser radar can receive control of a vehicle domain controller and a timing signal. In order to ensure the stability of the signal, the laser radar can be connected to the network router port in the vehicle by using a network cable. Then, the laser radar can be connected to the network router port in the vehicle by using a network cable. Figure 3 The laser radar-image 3D chessboard calibration board is used for joint calibration of the laser radar and the image sensor, so as to obtain a coordinate transformation matrix of the laser radar coordinate system converted to the image sensor coordinate system.

[0035] Further, in order to ensure the effectiveness of the subsequent positioning accuracy evaluation parameter calculation, the time when the laser radar starts to collect data and the time when the image sensor starts to collect data can be unified. Specifically, the laser radar can be started first, and then the laser radar and the image sensor can be uniformly timed by using the network router, so as to synchronize the data collection time of the laser radar and the image sensor.

[0036] Further, laser radar data collected by the laser radar can be obtained, and then laser mapping is performed by using the laser radar data to obtain laser mapping data. For example, the laser radar data can be input into mapping software (the specific type can not be limited) to perform laser mapping by using the mapping software.

[0037] S120: obtaining visual mapping data, the visual mapping data comprising a visual point cloud map.

[0038] In the embodiment of the application, the visual mapping data represents data obtained after visual mapping by using image sensor data. In an implementation manner, the visual mapping data can comprise a visual point cloud map (i.e., a map composed of point cloud data obtained by scanning by using an image sensor).

[0039] As an implementation manner, the visual mapping data can be obtained by using an image sensor. The image sensor in the embodiment of the application can be a vehicle-mounted sensor platform, which can specifically comprise an image camera, an IMU (Inertial Measurement Unit, inertial measurement unit), and the like. Optionally, the specific type and installation position of the image camera can not be limited. For example, the image camera can be a monocular camera, a binocular camera, or a combination of the two, and the image camera can be installed at the front of the vehicle (such as the front of the vehicle), at the rear of the vehicle (such as the trunk of the vehicle), or at the bottom of the vehicle.

[0040] As an implementation manner, in the case that the laser radar and the vehicle-mounted sensor platform have been synchronized by timing, vehicle-mounted sensor platform data (including image video data and IMU data, and the like) can be obtained, and then visual mapping is performed by using the vehicle-mounted sensor platform data to obtain the visual mapping data.

[0041] It can be understood that although the starting time of the laser radar has been unified with the starting time of the vehicle sensor platform, the laser radar data collected by the laser radar and the vehicle sensor platform data collected by the vehicle sensor platform are still different types of data. In order to improve the effectiveness of the comparison between the data, the laser mapping data can be first converted into second visual mapping data (for the sake of distinction, the visual mapping data converted from the laser radar data is referred to as second visual mapping data) through the aforementioned coordinate transformation matrix, and then the visual mapping data and the second visual mapping data are aligned in the same coordinate system. For example, the visual mapping data and the second visual mapping data can be aligned in the coordinate system in which the visual mapping data is located.

[0042] It is worth noting that the laser point cloud map and the visual point cloud map after alignment are located in the same coordinate system.

[0043] S130: Obtain a first error parameter between the laser point cloud map and the visual point cloud map, and obtain a second error parameter between the laser point cloud map and the visual point cloud map. The point cloud scale corresponding to the first error parameter is larger than the point cloud scale corresponding to the second error parameter.

[0044] In the embodiment of the application, the first error parameter represents the average vertical projection distance error, and the second error parameter represents the average distance error. The point cloud scale corresponding to the first error parameter is larger than the point cloud scale corresponding to the second error parameter. That is, the first error parameter is calculated based on large-scale point cloud features, and the second error parameter is calculated based on small-scale (relative to the first error parameter) point cloud features. Optionally, the scale here can be understood as the size, and the point cloud scale represents the size of the region occupied by the object composed of the point cloud in the point cloud map.

[0045] Since the laser point cloud map inevitably has mapping errors in the formation process, and its accuracy is more easily affected by the scene environment, in order to reduce or avoid the influence of mapping errors or scene environment on the positioning accuracy evaluation, the first error parameter between the laser point cloud map and the visual point cloud map can be obtained, and the second error parameter between the laser point cloud map and the visual point cloud map can be obtained, so as to evaluate the positioning accuracy of the output result of the preset visual mapping algorithm by using the two error parameters at the same time.

[0046] In the embodiments of the present application, the output result of the preset visual mapping algorithm includes laser mapping data and visual mapping data. The laser mapping data includes a laser point cloud map, and the visual mapping data includes a visual point cloud map. Since the positioning function of the intelligent parking task is mainly realized by relying on the map output by the preset visual mapping algorithm, by obtaining the error parameters in two different scales between the laser point cloud map and the visual point cloud map, the positioning accuracy of the output result of the preset visual mapping algorithm can be directly and objectively evaluated.

[0047] Referring to Figure 4 As an embodiment, step S130 can include:

[0048] S131: Obtain at least one structured region in the laser point cloud map.

[0049] In the process of obtaining the first error parameter between the laser point cloud map and the visual point cloud map, as an embodiment, at least one structured region in the laser point cloud map can be obtained first. The structured region represents a region surrounded by point clouds at positions of objects with fixed shapes. Optionally, the objects with fixed shapes can include large-scale objects such as ground, wall, and / or ceiling. Specifically, the objects with fixed shapes can be relatively large-scale objects such as ground, wall, or ceiling in an indoor parking lot environment. In the embodiments of the present application, the objects with fixed shapes can be determined (or set) in advance. Optionally, in some other embodiments, the structured region can also be a region surrounded by point clouds at positions of objects with non-fixed shapes, that is, the shape type of the structured region can not be limited, and can be regular or irregular.

[0050] Optionally, the size of the wall can be understood as the point cloud scale corresponding to the first error parameter. The size of the wall can be understood as the size of the region occupied by the wall formed by point clouds in the point cloud map.

[0051] The number of structured regions can be at least one. The objects represented by different structured regions can be the same, for example, they can all be ceilings. The objects represented by different structured regions can also be different, for example, the object represented by one structured region can be a ceiling, and the object represented by another structured region can be a ground.

[0052] The mapping evaluation (i.e., positioning accuracy evaluation) in the embodiments of the present application is mainly for large indoor parking lot environment, and the main objects with fixed shapes in the scene (such as the aforementioned ground, wall, or ceiling, etc.) have an impact on the evaluation of the mapping result in the large indoor parking lot. In order to accurately obtain at least one structured region in the laser point cloud map, as an implementation manner, at least one structured region in the laser point cloud map can be extracted by means of plane detection and segmentation.

[0053] S132: Obtain at least one visual point cloud data corresponding to the at least one structured region in the visual point cloud map.

[0054] Since the laser point cloud map and the visual point cloud map are in the same coordinate system, in the case of extracting at least one structured region of the laser point cloud map, in order to obtain objective and effective evaluation data, the point cloud data of the corresponding structured region of the visual point cloud map can be obtained based on the at least one structured region, so as to obtain at least one visual point cloud data. That is, for each structured region of the extracted laser point cloud map, the structured region includes both laser point cloud and visual point cloud.

[0055] S133: For each visual point cloud data, obtain part of the point cloud data in the visual point cloud data as reference visual point cloud.

[0056] In order to reduce the amount of data processing, as an implementation manner, for each visual point cloud data, a specified number of visual point clouds can be randomly extracted from the visual point cloud data as reference visual point cloud. Optionally, the specific value of the specified number can not be limited.

[0057] The reference visual point cloud is a point set, and the point set includes at least one point.

[0058] S134: Determine the average vertical projection distance error of the reference visual point cloud to the fitting plane where the structured region is located as the first error parameter.

[0059] As an implementation manner, the vertical projection distance of each point in the reference visual point cloud to the fitting plane where the structured region is located can be obtained, and then the average vertical projection distance error of the reference visual point cloud to the fitting plane where the structured region is located is determined as the first error parameter according to the following formula (1).

[0060]

[0061] wherein Error1 represents the first error parameter, N represents the number of points in the reference visual point cloud, i represents the i-th point in the reference visual point cloud, Pesti,i represents the i-th point in the visual mapping given plane region point cloud, Pproj,i represents the projection point of the i-th point in the visual mapping given plane region point cloud on the plane detected in the laser point cloud, is the square of the vector length, and ∑ represents the summation of all points in the visual mapping given plane region point cloud.

[0062] In one specific application scenario, please refer to Figure 5 , which shows the schematic flowchart of acquiring the first error parameter between the laser point cloud map and the visual point cloud map provided by the embodiments of the present application. As Figure 5 indicated, the structured region can be extracted from the laser map data (i.e. the laser point cloud map) in the form of plane detection Figure 6 and segmentation, and then the visual point cloud Figure 7 of the same structured region in the visual map data (i.e. the visual point cloud map) can be extracted, and then for the visual point cloud of any one of the extracted structured regions, a point set including multiple points can be randomly selected from the visual point cloud, and then the vertical projection distance of each point in the point set to the plane where the corresponding structured region is located can be calculated to obtain multiple vertical projection distances, and then the multiple vertical projection distances are averaged according to the above formula to obtain the average vertical projection distance error as the first error parameter.

[0063] Please refer to Figure 8 , as another embodiment, step S130 can include:

[0064] S135: Acquire the first target point cloud in the laser point cloud map.

[0065] wherein the first target point cloud is the point cloud in the laser point cloud map for representing a given target object. The given target object is distributed in a local region of the laser point cloud map. For example, the given target object can be a signboard, a fire hydrant, a gate, a passageway entrance and exit, and other objects with relatively small scales in an indoor parking lot environment.

[0066] In the process of obtaining the second error parameter between the laser point cloud map and the visual point cloud map, as a manner, a first target point cloud in the laser point cloud map can be obtained first. As an implementation manner, a preset semantic target can be obtained first, and then a point cloud representing the preset semantic target in the laser point cloud map is obtained as the first target point cloud. Specifically, a target object (i.e., the point cloud representing the preset semantic target) in the laser point cloud map can be detected by a target detection manner, and then the detected target object is segmented from the laser point cloud map to obtain the first target point cloud.

[0067] Optionally, in order to quickly detect the target object in the laser point cloud map, a semantic identifier can be set for the target object to realize the quick detection of the first target point cloud by extracting a given semantic target (i.e., the target object including the semantic identifier).

[0068] S136: Obtain a second target point cloud in the visual point cloud map, the first target point cloud and the second target point cloud having the same semantic.

[0069] The second target point cloud is a point cloud in the visual point cloud map for representing a given target object. The first target point cloud and the second target point cloud in the embodiment of the application have the same semantic, that is, the target object used for detecting the first target point cloud and the target object type used for detecting the second target point cloud are the same.

[0070] As a manner, while obtaining the first target point cloud in the laser point cloud map, the second target point cloud in the visual point cloud map can be extracted by a similar manner, for example, a point cloud representing a preset semantic target in the visual point cloud map can be obtained as the second target point cloud. The specific obtaining principle and implementation manner can be referred to the corresponding description in the foregoing implementation manner, and will not be described here.

[0071] S137: Obtain a first three-dimensional bounding box corresponding to the first target point cloud, and obtain a second three-dimensional bounding box corresponding to the second target point cloud.

[0072] In order to obtain objective and quantitative evaluation parameters, as a manner, the three-dimensional bounding box of the first target point cloud and the three-dimensional bounding box of the second target point cloud can be extracted respectively to obtain the first three-dimensional bounding box corresponding to the first target point cloud and the second three-dimensional bounding box corresponding to the second target point cloud.

[0073] Specifically, taking obtaining the first three-dimensional bounding box corresponding to the first target point cloud as an example, as shown in FIG. 6, the first three-dimensional bounding box corresponding to the first target point cloud can be obtained by extracting the three-dimensional bounding box of the first target point cloud. Figure 9As shown, since the first target point cloud includes the point cloud of the position where the given target object is located, that is, the first target point cloud includes a plurality of points, at this time, the minimum three-dimensional bounding box capable of enclosing the first target point cloud can be used as the first three-dimensional bounding box corresponding to the first target point cloud; similarly, the minimum three-dimensional bounding box capable of enclosing the second target point cloud can be used as the second three-dimensional bounding box corresponding to the second target point cloud.

[0074] S138: Determine the average distance error between the first three-dimensional bounding box and the second three-dimensional bounding box as the second error parameter.

[0075] As an implementation manner, the average distance error between the first three-dimensional bounding box and the second three-dimensional bounding box can be determined as the second error parameter according to the following average distance error calculation formula.

[0076]

[0077] Wherein, Error2 represents the second error parameter, "Cesti,i" represents the i-th corner point of the 3d bounding box of a certain visual semantic target point cloud given by visual mapping, "Cg,i" represents the i-th corner point of the 3d bounding box of a certain laser semantic target point cloud given by laser mapping, is the square of the length of the vector, and ∑ represents the summation of all corner points of the 3d bounding box of the semantic target point cloud.

[0078] In a specific application scenario, please refer to Figure 10 , which shows the schematic flow chart of obtaining the second error parameter between the laser point cloud map and the visual point cloud map provided by the embodiments of the present application. As Figure 10 and Figure 11 shown, target detection and segmentation can be performed on laser map data (i.e. laser point cloud map) and visual map data (i.e. visual point cloud map) respectively, to extract the given semantic target in the laser point cloud map (i.e. the laser map shown in Figure 10 ) and the given semantic target in the visual point cloud map (i.e. the visual map shown in Figure 10 ), and then the three-dimensional bounding box corresponding to the given semantic target in the laser point cloud map and the three-dimensional bounding box corresponding to the given semantic target in the visual point cloud map can be extracted respectively, and then the average distance error of the bounding box corner points based on the three-dimensional bounding boxes of the two is calculated as the second error parameter.

[0079] Wherein, Figure 11 The laser point cloud and the laser semantic target point cloud shown represent the first target point cloud, the visual point cloud and the visual semantic target point cloud represent the second target point cloud, and the 3d bounding box is the three-dimensional bounding box.

[0080] S140: Evaluate the positioning accuracy of the preset visual mapping algorithm based on the first error parameter and the second error parameter.

[0081] In evaluating the positioning accuracy of a preset visual mapping algorithm based on a first error parameter and a second error parameter, to more accurately assess the positioning accuracy of the output map, one approach is to define the sum of the first and second error parameters obtained from the output map of the preset visual mapping algorithm as the positioning accuracy evaluation parameter of the algorithm's output. A threshold value for this evaluation parameter can then be set. In this approach, if the positioning accuracy evaluation parameter is less than the threshold, the positioning accuracy of the preset visual mapping algorithm's output is considered high; conversely, if the positioning accuracy evaluation parameter is greater than or equal to the threshold, the positioning accuracy is considered low. The specific value of the threshold value can be set according to actual needs and is not limited in any particular case.

[0082] As another approach, the smaller the value of the positioning accuracy evaluation parameter, the higher the positioning accuracy of the output result of the preset visual mapping algorithm; conversely, the larger the value of the positioning accuracy evaluation parameter, the lower the positioning accuracy of the output result of the preset visual mapping algorithm.

[0083] This embodiment provides a positioning accuracy evaluation method. It acquires laser mapping data, including a laser point cloud map; acquires visual mapping data, including a visual point cloud map; acquires a first error parameter between the laser point cloud map and the visual point cloud map; and acquires a second error parameter between the two maps. The point cloud scale corresponding to the first error parameter is larger than the point cloud scale corresponding to the second error parameter. The point cloud scale characterizes the size of the area occupied by objects formed by the point cloud in the point cloud map. The positioning accuracy of a preset visual mapping algorithm is evaluated based on the first and second error parameters. By acquiring errors within two different scale ranges between the laser point cloud map and the visual point cloud map, and then evaluating the positioning accuracy of the preset visual mapping algorithm based on the sum of these two errors, the method allows for direct evaluation of the visual mapping algorithm's output based on the map data used for positioning, improving the effectiveness of positioning accuracy evaluation. Furthermore, since expensive high-precision sensors are not introduced to acquire data, the cost of positioning accuracy evaluation can be reduced.

[0084] Please see Figure 12 This application provides another embodiment of a positioning accuracy evaluation method, the method comprising:

[0085] S211: Obtain laser mapping data, the laser mapping data comprising a laser point cloud map.

[0086] The specific implementation of S211 can refer to the related description of S110 in the foregoing embodiments, and details are not described herein again.

[0087] S221: Obtain visual mapping data, the visual mapping data comprising a visual point cloud map.

[0088] The specific implementation of S221 can refer to the related description of S120 in the foregoing embodiments, and details are not described herein again.

[0089] S231: Obtain a first error parameter between the laser point cloud map and the visual point cloud map, and obtain a second error parameter between the laser point cloud map and the visual point cloud map, the point cloud scale corresponding to the first error parameter being greater than the point cloud scale corresponding to the second error parameter.

[0090] The specific implementation of S231 can refer to the related description of S130 in the foregoing embodiments, and details are not described herein again.

[0091] S241: Evaluate the positioning accuracy of a preset visual mapping algorithm according to the first error parameter and the second error parameter.

[0092] The specific implementation of S241 can refer to the related description of S140 in the foregoing embodiments, and details are not described herein again.

[0093] S212: Obtain adjacent frames with a time interval greater than a specified time interval and a distance interval greater than a specified distance interval in the laser mapping trajectory as first reference frames.

[0094] The laser mapping data in the embodiments of the present application can also comprise a laser mapping trajectory, and the visual mapping data can also comprise a visual mapping trajectory. In order to further objectively and quantitatively evaluate the output result of the preset visual mapping algorithm, in the process of executing steps S211-S241, evaluation indexes can also be obtained from the level of the laser mapping trajectory and the visual mapping trajectory to more comprehensively evaluate the positioning accuracy of the output result of the preset visual mapping algorithm.

[0095] In the process of obtaining evaluation metrics from both laser mapping trajectories and visual mapping trajectories, due to the differences between LiDAR and vehicle-mounted sensing platforms—such as hardware installation errors, non-rigid hardware connection errors during platform movement, synchronization errors in software algorithms, and calibration errors—it is necessary to eliminate the impact of these inherent errors on the subsequent acquisition of evaluation metrics. For laser mapping trajectories and visual mapping trajectories acquired and calculated within the same time period, a time and distance filter should be applied first. Taking laser mapping trajectories as an example, adjacent frames within the laser mapping trajectory whose time interval and distance interval are both greater than a specified time interval can be selected as the first reference frames.

[0096] The specified time interval can be set to the average synchronization time error between the LiDAR and the vehicle-mounted sensing platform, and the specified distance interval can be set to the average calibration translation distance error between the LiDAR and the vehicle-mounted sensing platform. The values ​​of the specified time interval and the specified distance interval can be obtained during the time synchronization process between the LiDAR and the vehicle-mounted sensing platform. The specific acquisition principle and implementation method will not be elaborated here.

[0097] As a specific implementation method, in the process of acquiring adjacent frames in the laser mapping trajectory where the time interval is greater than a specified time interval and the distance interval is greater than a specified distance interval, the average synchronization time error can be used as a threshold, and the frames can be resampled at 2 to 3 times the frame rate to perform the first round of screening of the laser mapping trajectory; then, the average calibration translation distance error can be used as a threshold, and the frames can be filtered at 2 to 3 times the distance to perform the second round of screening of the results after the first round of screening; finally, the obtained frames are used as the first reference frames.

[0098] In a specific application scenario, please refer to Figure 13 This document illustrates an example flowchart of an embodiment of the present application for obtaining adjacent frames in a laser mapping trajectory whose time interval is greater than a specified time interval and whose distance interval is greater than a specified distance interval. Figure 13 As shown, for a laser mapping trajectory, the pose data of its i-th frame can be extracted, and then the time interval between the pose data of its i-th frame and the pose data of its (i-1)-th frame can be obtained. Then, it is determined whether this time interval is greater than a specified time interval. Optionally, if the time interval is greater than (or equal to) the specified time interval, the distance interval between the pose data of its i-th frame and the pose data of its (i-1)-th frame can be obtained, and then it is determined whether this distance interval is greater than a specified distance interval; if the time interval is less than the specified time interval, the pose data of the i-th frame can be obtained again. Optionally, if the distance interval is greater than (or equal to) the specified distance interval, the filtered pose data can be inserted into the filtered trajectory data (i.e., the filtered pose data is used as the first reference frame); and if the distance interval is less than the specified distance interval, the pose data of the i-th frame can be obtained again.

[0099] S222: Obtain the adjacent frames with time intervals greater than the specified time interval and distance intervals greater than the specified distance interval in the visual mapping trajectory as the second reference frames.

[0100] Similarly, for the visual mapping trajectory, the adjacent frames with time intervals greater than the specified time interval and distance intervals greater than the specified distance interval in the visual mapping trajectory can be obtained as the second reference frames in the above manner. The acquisition process of the second reference frames is similar to that of the first reference frames, which will not be described here.

[0101] S232: Obtain the corresponding absolute trajectory error and relative trajectory error based on the first reference frames and the second reference frames.

[0102] The first reference frames and the second reference frames in the embodiments of the present application are both a set, that is, the first reference frames include multiple frames of poses, and the second reference frames also include multiple frames of poses. In the case where the first reference frames and the second reference frames are obtained, the corresponding absolute trajectory error can be obtained based on the first reference frames and the second reference frames according to the following formula (3), and the corresponding relative trajectory error can be obtained based on the first reference frames and the second reference frames according to the following formula (4).

[0103]

[0104] wherein, is the i-th pose matrix in the real trajectory given by the laser mapping, Testi,i is the corresponding i-th pose matrix in the estimated trajectory given by the visual mapping, is the square of the matrix norm, and ∑ represents the summation of the sampled pose points.

[0105]

[0106] wherein, is the i-th pose matrix in the real trajectory given by the laser mapping, T g,i +Δt represents the pose matrix at a fixed time interval from the i-th pose in the real trajectory, T esti,i is the corresponding i-th pose matrix in the estimated trajectory given by the visual mapping, T esti,i +Δt represents the pose matrix at a fixed time interval from the i-th pose in the estimated trajectory, is the square of the matrix norm, and ∑ represents the summation of the sampled pose points.

[0107] Specifically, in order to verify the effectiveness and stability of the data, multiple absolute trajectory errors can be calculated by substituting the sampled frames from the first reference frames and the second reference frames into the absolute trajectory error calculation formula, respectively. The curve graph of the multiple absolute trajectory errors is as shown in FIG. 6. Figure 14and the second reference frame, and the relative trajectory errors are calculated (a curve diagram of the relative trajectory errors is shown in FIG. 6B). The values of the calculated absolute trajectory errors are different, and thus the maximum value, the minimum value, the average value, and the median value of the absolute trajectory errors can be calculated. Similarly, the values of the calculated relative trajectory errors are also different, and thus the maximum value, the minimum value, the average value, and the median value of the relative trajectory errors can be calculated. Figure 15 The values of the calculated absolute trajectory errors are different, and thus the maximum value, the minimum value, the average value, and the median value of the absolute trajectory errors can be calculated. Similarly, the values of the calculated relative trajectory errors are also different, and thus the maximum value, the minimum value, the average value, and the median value of the relative trajectory errors can be calculated.

[0108] It should be noted that, in the process of calculating the absolute trajectory error and the relative trajectory error, the laser mapping trajectory represents the real trajectory, and the visual mapping trajectory represents the measured trajectory.

[0109] S242: Evaluate the positioning accuracy of the preset visual mapping algorithm based on the absolute trajectory error, the relative trajectory error, the first error parameter, and the second error parameter.

[0110] In the embodiment, the positioning accuracy of the output result of the preset visual mapping algorithm can be comprehensively evaluated based on the absolute trajectory error and the relative trajectory error, and the aforementioned positioning accuracy evaluation parameters (including the first error parameter and the second error parameter), so that the positioning accuracy of the output result of the visual mapping can be evaluated based on the trajectory output result and the map output result of the preset visual mapping algorithm at the same time. Compared with evaluating the positioning accuracy of the output result of the visual mapping only by one evaluation index or by the trajectory output result of the visual mapping algorithm, the positioning accuracy evaluation method provided in the embodiment can more comprehensively and accurately evaluate the positioning accuracy of the output result of the visual mapping, so as to improve the effectiveness and reliability of the positioning accuracy evaluation.

[0111] Optionally, the results of the absolute trajectory error and the relative trajectory error can more truly reflect the real trajectory estimation error obtained by different sensors and different algorithm schemes, and thus when the positioning accuracy of the preset visual mapping algorithm is evaluated based on the absolute trajectory error and the relative trajectory error, and the positioning accuracy evaluation parameters, the relationship between the evaluation index and the positioning accuracy can not be limited, and it can be understood as an open positioning accuracy evaluation method.

[0112] In some other embodiments, the mapping result can also be indirectly evaluated by comparing the trajectory output by the high-precision combined navigation. Or the mapping result can be evaluated by the high-precision map data provided by an external third-party map provider. The specific evaluation method can be determined according to actual needs or requirements for the positioning accuracy.

[0113] The positioning accuracy evaluation method provided in the embodiment can directly evaluate the output result of the visual mapping algorithm according to the map data used for positioning, and improve the effectiveness of the positioning accuracy evaluation. Meanwhile, since the expensive high-precision sensor is not introduced to obtain data, the evaluation cost of the positioning accuracy can be reduced.

[0114] Further, by obtaining the absolute trajectory error and the relative trajectory error between the laser mapping trajectory and the visual mapping trajectory, and then comprehensively evaluating the positioning accuracy of the output result of the preset visual mapping algorithm based on the absolute trajectory error, the relative trajectory error and the aforementioned positioning accuracy evaluation parameter, the positioning accuracy of the output result of the visual mapping can be evaluated based on the trajectory output result and the map output result of the preset visual mapping algorithm. Compared with evaluating the positioning accuracy of the output result of the visual mapping only by one evaluation index or by the trajectory output result of the visual mapping algorithm, the positioning accuracy of the output result of the visual mapping can be more comprehensively and accurately evaluated by the positioning accuracy evaluation method provided in the embodiment, so that the effectiveness and reliability of the positioning accuracy evaluation can be improved.

[0115] Please refer to Figure 16 The positioning accuracy evaluation device 300 provided in the present application comprises:

[0116] The laser mapping result acquisition module 310 is configured to acquire laser mapping data, wherein the laser mapping data comprises a laser point cloud map.

[0117] As an implementation manner, the laser mapping result acquisition module 310 can be specifically configured to acquire the laser mapping data by a reference sensor, wherein the reference sensor comprises a laser radar.

[0118] The visual mapping result acquisition module 320 is configured to acquire visual mapping data, wherein the visual mapping data comprises a visual point cloud map.

[0119] As an implementation manner, the visual mapping result acquisition module 320 can be specifically configured to acquire the visual mapping data by an image sensor, wherein the image sensor comprises a vehicle-mounted sensor platform.

[0120] The error parameter acquisition module 330 is configured to acquire a first error parameter between the laser point cloud map and the visual point cloud map, and acquire a second error parameter between the laser point cloud map and the visual point cloud map. The first error parameter corresponds to a point cloud scale larger than the second error parameter. The point cloud scale represents the size of an area occupied by an object formed by point cloud in the point cloud map.

[0121] As an implementation, the error parameter acquisition module 330 can be specifically configured to acquire at least one structured area in the laser point cloud map; acquire at least one piece of visual point cloud data in the visual point cloud map corresponding to the at least one structured area; for each piece of visual point cloud data, acquire part of the point cloud data from the visual point cloud data as a reference visual point cloud; and determine an average vertical projection distance error of the reference visual point cloud to a fitting plane where the structured area is located as the first error parameter.

[0122] As an implementation, the acquisition of the at least one structured area in the laser point cloud map can include: acquiring the at least one structured area in the laser point cloud map by means of plane detection.

[0123] As another implementation, the error parameter acquisition module 330 can be specifically configured to acquire a first target point cloud in the laser point cloud map; acquire a second target point cloud in the visual point cloud map, the first target point cloud and the second target point cloud having the same semantics; acquire a first three-dimensional bounding box corresponding to the first target point cloud, and acquire a second three-dimensional bounding box corresponding to the second target point cloud; and determine an average distance error between the first three-dimensional bounding box and the second three-dimensional bounding box as the second error parameter.

[0124] As an implementation, the acquisition of the first target point cloud in the laser point cloud map can include: acquiring a preset semantic target; and acquiring point cloud representing the preset semantic target in the laser point cloud map as the first target point cloud. Similarly, the acquisition of the second target point cloud in the visual point cloud map can include: acquiring point cloud representing the preset semantic target in the visual point cloud map as the second target point cloud.

[0125] The positioning accuracy evaluation module 340 is configured to evaluate the positioning accuracy of a preset visual mapping algorithm according to the first error parameter and the second error parameter.

[0126] In this embodiment, the positioning accuracy evaluation device 300 can further include an evaluation parameter acquisition module configured to acquire, as a first reference frame, adjacent frames in the laser mapping trajectory having a time interval greater than a specified time interval and a distance interval greater than a specified distance interval; acquire, as a second reference frame, adjacent frames in the visual mapping trajectory having a time interval greater than a specified time interval and a distance interval greater than a specified distance interval; and acquire corresponding absolute trajectory errors and relative trajectory errors based on the first reference frame and the second reference frame. In this way, the positioning accuracy evaluation module 340 can be specifically configured to evaluate the positioning accuracy of a preset visual mapping algorithm based on the absolute trajectory errors, the relative trajectory errors, the first error parameter, and the second error parameter.

[0127] The above will be described in detail below. Figure 17 A vehicle provided by the present application will be described.

[0128] Please refer to Figure 17 Based on the above positioning accuracy evaluation method and device, the present embodiment further provides another vehicle 100 that can perform the foregoing positioning accuracy evaluation method. The vehicle 100 includes one or more (only one is shown in the figure) processors 102, a memory 104, and a data acquisition module 106 coupled with each other. The memory 104 stores a program that can perform the content of the foregoing embodiments, and the processor 102 can execute the program stored in the memory 104.

[0129] The processor 102 can include one or more processing cores. The processor 102 connects various parts in the vehicle 100 through various interfaces and lines, performs various functions of the vehicle 100 and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 104, and calling data stored in the memory 104. Optionally, the processor 102 can be implemented in at least one of a hardware form of a neural network processing unit (NPU), a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 102 can be integrated with a combination of one or more of a central processing unit (CPU), a graphics processing unit (GPU), a neural network processing unit (NPU), and a modem. Among them, the CPU mainly processes operating systems, user interfaces, and application programs; the GPU is responsible for rendering and drawing display content; the NPU is responsible for processing multimedia data such as video and images; and the modem is used for processing wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor 102, but can be implemented separately through a communication chip.

[0130] The memory 104 can include a random access memory (RAM) and can also include a read-only memory (ROM). The memory 104 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 104 can include a program storage area and a data storage area, wherein the program storage area can store instructions for implementing an operating system, instructions for implementing at least one function (such as a touch function, a sound playing function, an image playing function, etc.), instructions for implementing each of the following method embodiments, etc. The data storage area can also store data created by the vehicle 100 in use (such as a phonebook, audio and video data, chat record data), etc.

[0131] The data acquisition module 106 can be used to obtain driving state information of the vehicle 100, and the data acquisition module 106 can be a laser radar, a camera, a sensor, a GPS (Global Positioning System) navigation, etc.

[0132] The embodiment of the present application provides a computer readable storage medium. The computer readable storage medium stores program codes, and the program codes can be invoked by a processor to execute the method described in the method embodiment.

[0133] The computer readable storage medium can be an electronic storage such as a flash memory, an EEPROM (electrically erasable programmable read-only memory), an EPROM, a hard disk or a ROM. Alternatively, the computer readable storage medium comprises a non-volatile computer readable storage medium. The computer readable storage medium has a storage space for storing program codes for executing any method steps in the above method. The program codes can be read from or written into one or more computer program products. The program codes can be compressed in a suitable form, for example.

[0134] To sum up, the positioning accuracy evaluation method and device and vehicle provided by the present application can obtain laser mapping data, the laser mapping data comprising a laser point cloud map; obtain visual mapping data, the visual mapping data comprising a visual point cloud map; obtain a first error parameter between the laser point cloud map and the visual point cloud map, and obtain a second error parameter between the laser point cloud map and the visual point cloud map, the point cloud scale corresponding to the first error parameter being greater than the point cloud scale corresponding to the second error parameter, the point cloud scale representing the size of the region occupied by the object composed of the point cloud in the point cloud map; and evaluate the positioning accuracy of the preset visual mapping algorithm according to the first error parameter and the second error parameter. Thus, by obtaining the errors in two different scale ranges between the laser point cloud map and the visual point cloud map, and then evaluating the positioning accuracy of the preset visual mapping algorithm based on the sum of the two errors, the output result of the visual mapping algorithm can be directly evaluated according to the map data used for positioning, and the effectiveness of the positioning accuracy evaluation is improved. At the same time, since the data is not obtained by introducing an expensive high-precision sensor, the evaluation cost of the positioning accuracy can be reduced.

[0135] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement to some technical features thereof; and these modifications or replacements do not drive the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for evaluating positioning accuracy, characterized in that, The method includes: Acquire laser mapping data, which includes a laser point cloud map; Acquire visual mapping data, which includes a visual point cloud map; A first error parameter is obtained between the laser point cloud map and the visual point cloud map, and a second error parameter is obtained between the laser point cloud map and the visual point cloud map. In the laser point cloud map and the visual point cloud map, the objects are divided into large-scale objects and small-scale objects according to the size of the area occupied by the objects formed by the point cloud. The first error parameter is obtained based on the point cloud computing corresponding to the large-scale objects, and the second error parameter is obtained based on the point cloud computing corresponding to the small-scale objects. The positioning accuracy of the preset visual mapping algorithm is evaluated based on the first error parameter and the second error parameter.

2. The method according to claim 1, characterized in that, The laser mapping data and the visual mapping data are located in the same coordinate system. Obtaining the first error parameter between the laser point cloud map and the visual point cloud map includes: Obtain at least one structured region from the laser point cloud map; Obtain at least one visual point cloud data point in the visual point cloud map that corresponds to the at least one structured region; For each visual point cloud data, a portion of the point cloud data is obtained from the visual point cloud data as a reference visual point cloud; The average vertical projection distance error from the reference visual point cloud to the fitting plane containing the structured region is determined and used as the first error parameter.

3. The method according to claim 2, characterized in that, The step of obtaining at least one structured region in the laser point cloud map includes: At least one structured region in the laser point cloud map is obtained by planar detection.

4. The method according to claim 2, characterized in that, The step of obtaining the second error parameter between the laser point cloud map and the visual point cloud map includes: Obtain the first target point cloud in the laser point cloud map; Obtain the second target point cloud from the visual point cloud map, where the first target point cloud and the second target point cloud have the same semantics; Obtain the first 3D bounding box corresponding to the first target point cloud, and obtain the second 3D bounding box corresponding to the second target point cloud; The average distance error between the first 3D bounding box and the second 3D bounding box is determined as the second error parameter.

5. The method according to claim 4, characterized in that, The step of obtaining the first target point cloud in the laser point cloud map includes: Obtain the preset semantic target; Obtain the point cloud representing the preset semantic target in the laser point cloud map, and use it as the first target point cloud; The step of obtaining the second target point cloud in the visual point cloud map includes: Obtain the point cloud representing the preset semantic target in the visual point cloud map, and use it as the second target point cloud.

6. The method according to claim 1, characterized in that, The laser mapping data further includes a laser mapping trajectory, the visual mapping data further includes a visual mapping trajectory, and the method further includes: The adjacent frames in the laser mapping trajectory whose time interval is greater than a specified time interval and whose distance interval is greater than a specified distance interval are obtained and used as the first reference frames. Obtain adjacent frames in the visual mapping trajectory whose time interval is greater than a specified time interval and whose distance interval is greater than a specified distance interval, and use them as second reference frames; Based on the first reference frame and the second reference frame, obtain the corresponding absolute trajectory error and relative trajectory error; The method further includes: The positioning accuracy of the preset visual mapping algorithm is evaluated based on the absolute trajectory error, relative trajectory error, the first error parameter, and the second error parameter.

7. A positioning accuracy evaluation device, characterized in that, The device includes: The laser mapping result acquisition module is used to acquire laser mapping data, which includes a laser point cloud map. The visual mapping result acquisition module is used to acquire visual mapping data, which includes a visual point cloud map. An error parameter acquisition module is used to acquire a first error parameter between the laser point cloud map and the visual point cloud map, and to acquire a second error parameter between the laser point cloud map and the visual point cloud map. In the laser point cloud map and the visual point cloud map, the objects are divided into large-scale objects and small-scale objects according to the size of the area occupied by the objects formed by the point cloud. The first error parameter is obtained based on the point cloud computing corresponding to the large-scale objects, and the second error parameter is obtained based on the point cloud computing corresponding to the small-scale objects. The positioning accuracy evaluation module is used to evaluate the positioning accuracy of the preset visual mapping algorithm based on the first error parameter and the second error parameter.

8. A vehicle, characterized in that, Includes one or more processors and memory; One or more programs are stored in the memory and configured to be executed by the one or more processors, the one or more programs being configured to perform the method of any one of claims 1-6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores program code, wherein the program code, when executed by a processor, performs the method according to any one of claims 1-6.

10. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1-6.

Images