Radar motion compensation effect evaluation method and device, storage medium and equipment

Abstract

The invention discloses a radar motion compensation effect evaluation method and device, a storage medium and equipment, and the method comprises the steps: carrying out the motion compensation of a target radar installed in a target robot during the control of the free motion of the target robot, obtaining the real-time point cloud data through the target radar, and carrying out the calculation of the real-time point cloud data; the real-time point cloud data is used for updating the environment point cloud data, and then a spline curve is used for fitting the environment point cloud data to obtain fitted environment data; carrying out sampling processing on the fitting environment data to obtain ideal environment point cloud data; calculating the score of the distance between each point in the real-time point cloud data and the closest point in the ideal environment point cloud data; and furthermore, the outliers caused by motion compensation can be more accurately found by utilizing the distance score, so that the effective evaluation of the motion compensation effect of the target radar is realized, and an evaluation result with higher accuracy is obtained.

Description

Technical Field

[0001] This application relates to the field of robot technology, and in particular, to a method, device, storage medium, and equipment for evaluating the effect of radar motion compensation. Background Art

[0002] With the progress of robot technology, the types of robots are becoming more and more diverse, and the functions are becoming more and more powerful. For example, floor-sweeping robots equipped with lidar or other service robots bring great convenience to people's lives. However, due to possible partial defects or installation defects in the lidar sensor itself, the overall performance of the robot may decline.

[0003] In response to this, currently, the method of point cloud motion distortion compensation is often used to detect possible defects in the radar of the robot. Moreover, in order to improve the overall performance of the robot, the effect of radar motion compensation can also be evaluated to help researchers continuously improve the compensation method. Currently, there are two types of methods for evaluating the effect of radar motion compensation in the prior art: one is the indirect positioning method, that is, during the mass production and offline process of the robot, a scenario is designed, and a ground truth system is installed in this scenario. The motion pose of the robot in this scenario is compared with the ground truth system, and the root mean square error or the Euclidean distance of relative time is used as the positioning evaluation standard. Another commonly used evaluation method is the matching score method. Different from the first method that uses the ground truth system, this method saves the environmental data collected in the first frame as the relative environmental ground truth. During the movement of the robot, motion distortion compensation is performed on the single-frame point cloud of the radar in real time, and the compensated point cloud is matched with the point cloud collected in the first frame, and the Euclidean distance of each point after matching is used as the score value, and then the de-distortion effect is evaluated through this score value. However, the accuracy of these two existing evaluation methods is not high, and they cannot timely reflect the possible problems in radar motion compensation. Summary of the Invention

[0004] The main purpose of the embodiments of this application is to provide a method, device, storage medium, and equipment for evaluating the effect of radar motion compensation, which can effectively improve the evaluation accuracy of the radar motion compensation effect installed on the robot.

[0005] The embodiments of this application provide a method for evaluating the effect of radar motion compensation, including:

[0006] During the process of controlling the target robot to move freely, perform motion compensation on the target radar installed in the target robot, and obtain real-time point cloud data through the target radar, so as to update the environmental point cloud data by using the real-time point cloud data;

[0007] Use a spline curve to fit the updated environmental point cloud data to obtain the fitted environmental data; and perform sampling processing on the fitted environmental data to obtain the ideal environmental point cloud data;

[0008] Calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data;

[0009] Use the distance score to evaluate the motion compensation effect of the target radar to obtain an evaluation result.

[0010] In a possible implementation, the obtaining of the real-time point cloud data by the target radar to update the environmental point cloud data using the real-time point cloud data includes:

[0011] Obtain the static pose of the target robot after free movement, and after the target robot stops, obtain the static point cloud data through the target radar;

[0012] Determine whether the static point cloud data is the first-frame point cloud data. If so, use the static point cloud data as the initial environmental point cloud data; if not, register the static point cloud data with the existing environmental point cloud data according to the static pose to update the environmental point cloud data; and so on until the coverage range of the environmental point cloud data meets the conditions to obtain the updated environmental point cloud data.

[0013] In a possible implementation, the using of a spline curve to fit the updated environmental point cloud data to obtain the fitted environmental data includes:

[0014] Use a spline curve to splice the discrete point cloud data in the updated environmental point cloud data to complete the smoothness modeling of the environmental data and obtain the integrated smooth environmental data as the fitted environmental data.

[0015] In a possible implementation, the performing of sampling processing on the fitted environmental data to obtain the ideal environmental point cloud data includes:

[0016] Uniformly select N sampling points from the fitted environmental data as the ideal environmental point cloud data; N is a positive integer greater than 1.

[0017] In a possible implementation, the calculating of the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data includes:

[0018] Calculate the Euclidean distance between each point in the real-time point cloud data and each point in the ideal environmental point cloud data, and determine the point with the closest distance between each point in the real-time point cloud data and the ideal environmental point cloud data according to the obtained Euclidean distance calculation results;

[0019] Using an exponential function and a preset distance threshold, calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data.

[0020] In a possible implementation, during the process of controlling the target robot to move freely, performing motion compensation on the target radar installed in the target robot includes:

[0021] During the process of controlling the target robot to move freely, use an odometer composed of an inertial measurement unit (IMU) and a wheel speedometer to perform motion compensation on the target radar installed in the target robot.

[0022] In a possible implementation, after using the distance score to evaluate the motion compensation effect of the target radar and obtaining an evaluation result, the method further includes:

[0023] Verify whether the evaluation result is valid by adding position noise and / or attitude noise to the odometer to obtain a verification result.

[0024] The embodiment of the present application further provides a device for evaluating the motion compensation effect of a radar, including:

[0025] An acquisition unit, configured to perform motion compensation on the target radar installed in the target robot during the process of controlling the target robot to move freely, and acquire real-time point cloud data through the target radar to update the environmental point cloud data using the real-time point cloud data;

[0026] A fitting unit, configured to fit the environmental point cloud data using a spline curve to obtain fitted environmental data; and perform sampling processing on the fitted environmental data to obtain ideal environmental point cloud data;

[0027] A calculation unit, configured to calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data;

[0028] An evaluation unit, configured to evaluate the motion compensation effect of the target radar using the distance score to obtain an evaluation result.

[0029] In a possible implementation, the acquisition unit includes:

[0030] An acquisition subunit, configured to acquire the static pose of the target robot after free movement, and after the target robot stops, acquire static point cloud data through the target radar;

[0031] A judgment subunit, configured to judge whether the static point cloud data is the first-frame point cloud data. If so, the static point cloud data is used as the initial environmental point cloud data. If not, according to the static pose, the static point cloud data is registered with the existing environmental point cloud data to update the environmental point cloud data. And so on, until the coverage range of the environmental point cloud data meets the conditions, and the updated environmental point cloud data is obtained.

[0032] In a possible implementation manner, the fitting unit is specifically configured to:

[0033] Use a spline curve to splice the discrete point cloud data in the updated environmental point cloud data, complete the smoothness modeling of the environmental data, and obtain the integrated smooth environmental data as the fitted environmental data.

[0034] In a possible implementation manner, the fitting unit is specifically configured to:

[0035] Uniformly select N sampling points from the fitted environmental data as the ideal environmental point cloud data; N is a positive integer greater than 1.

[0036] In a possible implementation manner, the calculation unit includes:

[0037] A first calculation subunit, configured to calculate the Euclidean distance between each point in the real-time point cloud data and each point in the ideal environmental point cloud data, and determine the point in the real-time point cloud data that is closest to the point in the ideal environmental point cloud data according to the obtained Euclidean distance calculation result;

[0038] A second calculation subunit, configured to calculate the distance score between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it by using an exponential function and a preset distance threshold.

[0039] In a possible implementation manner, the acquisition unit is specifically configured to:

[0040] During the process of controlling the target robot to move freely, use an odometer composed of an inertial measurement unit (IMU) and a wheel speedometer to perform motion compensation on the target radar installed in the target robot.

[0041] In a possible implementation manner, the device further includes:

[0042] A verification unit, configured to verify whether the evaluation result is valid by adding position noise and / or attitude noise to the odometer, and obtain a verification result.

[0043] An embodiment of the present application further provides a device for evaluating the radar motion compensation effect, including: a processor, a memory, and a system bus;

[0044] The processor and the memory are connected through the system bus;

[0045] The memory is used to store one or more programs, and the one or more programs include instructions, which, when executed by the processor, cause the processor to execute any of the implementation manners of the above radar motion compensation effect evaluation method.

[0046] An embodiment of the present application further provides a computer-readable storage medium, in which instructions are stored. When the instructions run on a terminal device, the terminal device is caused to execute any of the implementation manners of the above radar motion compensation effect evaluation method.

[0047] An embodiment of the present application further provides a computer program product, which, when running on a terminal device, causes the terminal device to execute any of the implementation manners of the above radar motion compensation effect evaluation method.

[0048] A method, device, storage medium and equipment for evaluating the radar motion compensation effect provided by an embodiment of the present application first perform motion compensation on a target radar installed in a target robot during the process of controlling the target robot to move freely, and obtain real-time point cloud data through the target radar to update the environmental point cloud data by using the real-time point cloud data. Then, a spline curve is used to fit the environmental point cloud data to obtain fitted environmental data; and the fitted environmental data is sampled to obtain ideal environmental point cloud data; then, the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data is calculated; furthermore, the distance score can be used to evaluate the motion compensation effect of the target radar to obtain an evaluation result.

[0049] It can be seen that when evaluating the motion compensation effect of the target radar in the present application, a spline curve is first used to fit the real-time point cloud data obtained by the target radar during the free movement of the target robot to obtain smooth fitted environmental data, and then ideal environmental point cloud data is sampled from it. Furthermore, the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data can be calculated to more accurately find the outlier points caused by motion compensation, so as to effectively evaluate the motion compensation effect of the target radar. Compared with the prior art in which the environmental data collected in the first frame is used as the relative environmental true value or the motion compensation effect is verified based on the matching score of the Euclidean distance (it is difficult to find outlier points), the accuracy of the evaluation result is higher. Description of the Drawings

[0050] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required for the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are some embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other accompanying drawings can also be obtained based on these drawings.

[0051] Figure 1 Schematic diagram of the absence of point cloud data when the environmental data collected in the first frame is used as the relative environmental ground truth provided by the embodiment of the present application;

[0052] Figure 2 Schematic flowchart of a method for evaluating the radar motion compensation effect provided by the embodiment of the present application;

[0053] Figure 3 Schematic diagram of the process of determining the ideal environmental point cloud data using a spline curve provided by the embodiment of the present application;

[0054] Figure 4 Example diagram of the fitting effect of the spline curve provided by the embodiment of the present application;

[0055] Figure 5 Example diagram of the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data provided by the embodiment of the present application;

[0056] Figure 6 Example diagram of the result of verifying the effectiveness of the evaluation result by adding position noise to the odometer provided by the embodiment of the present application;

[0057] Figure 7 Example diagram of the result of verifying the effectiveness of the evaluation result by adding attitude noise to the odometer provided by the embodiment of the present application;

[0058] Figure 8 Schematic diagram of the composition of a device for evaluating the radar motion compensation effect provided by the embodiment of the present application. Detailed implementation manners

[0059] With the continuous innovation of robot technology, robots can not only be widely used in industrial scenarios, but also many service robots (such as floor-sweeping robots) are rapidly popularized in people's home lives. Currently, most service robots applied to homes use their own sensors (such as lidar) to complete perception, positioning, and path planning. In this case, many challenges are also brought, including challenges of environmental variability and sensor (such as lidar) capabilities. Under the premise of limited sensor capabilities, there may be some defects or installation defects (such as installation angle errors) in sensors such as lidar itself, which will all lead to a decline in the overall performance of the robot.

[0060] Regarding the above installation errors and possible defects of the sensor itself, during the mass production and offline process of the robot, robot manufacturers usually set up detection schemes such as point cloud motion distortion compensation to detect possible defects in sensors such as the radar of the robot. Moreover, in order to improve the overall performance of the robot, the effect of radar motion compensation can also be evaluated to help researchers continuously improve the compensation method. Currently, the evaluation methods for the effect of point cloud de-distortion often evaluate through the matching score after point cloud registration, and some evaluate through the positioning effect, lacking generalization.

[0061] Specifically, the commonly used evaluation methods for the radar motion compensation effect in the existing technologies currently include the following two:

[0062] The first is the indirect positioning method.

[0063] During the mass production and offline process of the robot (marking that the robot has completed production and is ready for shipment), design a scenario in which a ground truth system is installed, and then compare the motion pose of the robot in this scenario with the ground truth system. Use the root mean square error or the Euclidean distance relative to time as the positioning evaluation criterion to evaluate the radar motion compensation effect of the robot.

[0064] The second is the matching score method.

[0065] This method builds the environment according to the first indirect positioning method, but does not use the ground truth system. Instead, it saves the environmental data collected in the first frame as the relative environmental ground truth. During the movement of the robot, perform real-time motion distortion compensation on the single-frame point cloud of the radar, and match the compensated point cloud with the point cloud collected in the first frame. Then use the Euclidean distance of the matched points as the score value to evaluate the de-distortion effect after motion compensation through this score value. That is, the smaller the score value, the closer the compensated point cloud is to the first-frame point cloud, and the better the de-distortion effect. As Figure 1 shown is the calculation method of the point cloud de-distortion score value under ideal conditions. Among them, point cloud P represents the environmental point cloud, and point cloud Q represents the situation where there are some missing parts in the compensated point cloud. For example, due to occlusion, etc., the point cloud information may be incomplete.

[0066] It can be seen that this method uses the first-frame map as the global ground truth map, which has the risk of incomplete point cloud information, resulting in the inability to fully represent the environment, thus affecting subsequent matching and score calculation, and causing the evaluation result to be inaccurate.

[0067] Moreover, both of the existing evaluation methods verify the radar motion compensation effect based on the matching scores of Euclidean distance, making it difficult to detect outliers, which are often caused by motion compensation. Additionally, the scoring calculation methods of the registration algorithms (such as the classic Iterative Closest Point Registration Algorithm (ICP)) used in the existing evaluation methods are often designed to reflect the function of partial overlap of point clouds, ignoring points with large distances or small score values (i.e., ignoring outliers), which will also reduce the accuracy of the evaluation method and thus cannot promptly reflect potential problems in radar motion compensation.

[0068] Therefore, there is an urgent need for a solution that can accurately detect the effect of removing distortion from the point cloud of a single-line lidar. Especially for low-cost single-line radars with a low frequency (such as 5Hz), the points with later timestamps in a single-frame point cloud often have a large deviation from the actual situation, but it is difficult to detect such problems relying on the existing positioning effects or matching score methods.

[0069] To address the above deficiencies, the present application provides a method for evaluating the radar motion compensation effect. First, during the process of controlling the target robot to move freely, motion compensation is performed on the target radar installed in the target robot, and real-time point cloud data is obtained through the target radar to update the environmental point cloud data using the real-time point cloud data. Then, a spline curve is used to fit the environmental point cloud data to obtain the fitted environmental data; and the fitted environmental data is sampled to obtain the ideal environmental point cloud data; Next, the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data is calculated; Furthermore, the distance score can be used to evaluate the motion compensation effect of the target radar to obtain an evaluation result.

[0070] It can be seen that when evaluating the radar motion compensation effect in the present application, a spline curve is first used to fit the real-time point cloud data obtained by the target radar during the free movement of the target robot to obtain smooth fitted environmental data, and then ideal environmental point cloud data is sampled from it. Furthermore, the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data can be calculated to more accurately detect outliers caused by motion compensation, thereby realizing an effective evaluation of the radar motion compensation effect. Compared with the prior art that uses the environmental data collected in the first frame as the relative environmental true value or verifies the motion compensation effect based on the matching scores of Euclidean distance (which is difficult to detect outliers), the accuracy of the evaluation result is higher.

[0071] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions in the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this application. Apparently, the described embodiments are some, but not all, of the embodiments of this application. All other embodiments obtained by those of ordinary skill in the art based on the embodiments in this application without making creative efforts shall fall within the scope of protection of this application.

[0072] First Embodiment

[0073] See Figure 2 , which is a schematic flowchart of a method for evaluating the radar motion compensation effect provided in this embodiment. The method includes the following steps:

[0074] S201: During the process of controlling the target robot to move freely, perform motion compensation on the target radar installed in the target robot, and obtain real-time point cloud data through the target radar to update the environmental point cloud data using the real-time point cloud data.

[0075] In this embodiment, any radar that needs to have its motion compensation effect evaluated is defined as the target radar, and the robot installed with the target radar is defined as the target robot. It should be noted that this embodiment does not limit the type of the target radar. For example, the target radar can be the motion of a low-cost single-line radar, etc.; at the same time, this embodiment also does not limit the type of the target robot. For example, the target robot can be a sweeping robot or other service robots, etc.

[0076] It should be noted that for the problem of low accuracy when using the environmental data collected in the first frame as the relative environmental ground truth or the existing matching scores based on Euclidean distance (such as the ICP calculation score) to verify the motion compensation effect (it is difficult to detect outliers), this application proposes to first use the form of interacting with the robot (such as voice, gesture, line of sight, etc.) and fitting with a spline curve to complete the smoothness modeling of the environmental data, and obtain the overall smooth real environmental point cloud data. Then, by calculating the distance score between each point in the real-time point cloud data and the point with the closest distance in the smooth real environmental point cloud data (i.e., the ideal environmental point cloud data mentioned later), more accurate outliers caused by motion compensation of the target radar can be found, so as to accurately evaluate the motion compensation effect of the target radar and timely feedback the problems existing in the motion compensation of the target radar.

[0077] Specifically, in order to improve the evaluation accuracy of the motion compensation effect of the target radar installed on the target robot, first, the target robot can be controlled to automatically rotate at different angles through interaction with the target robot (such as voice, gesture, line of sight and other interaction methods) to achieve the free movement of the target robot. Moreover, during the process of controlling the free movement of the target robot, motion compensation can be performed on the target radar installed in the target robot, and real-time point cloud data (after compensation) can be obtained through the target radar to update the environmental point cloud data using these real-time point cloud data for subsequent step S202.

[0078] Among them, an optional implementation method is that the implementation process of obtaining real-time point cloud data through the target radar to update the environmental point cloud data using the real-time point cloud data can specifically include: First, obtain the static pose of the target robot after free movement, and after the target robot stops, obtain the static point cloud data through the target radar. Then, determine whether the static point cloud data is the first-frame point cloud data. If so, use this static point cloud data as the initial environmental point cloud data; if not, register the static point cloud data with the existing environmental point cloud data according to the static pose to update the environmental point cloud data; and so on until the coverage range of the environmental point cloud data meets the conditions (the specific content is not limited, such as the point cloud covers more than 70% of the entire environment, etc.) to obtain the updated environmental point cloud data.

[0079] In this implementation method, as Figure 3 shown, first, the target robot can be controlled to move freely, and when the target robot stops after free movement, record the static pose (i.e., the relative pose, and represented as T ∈ SO(3)), and obtain the lidar data after stopping through the target radar (here it is defined as the static point cloud data).

[0080] Among them, the pose of the target robot refers to the position and orientation of the target robot in space. The relative pose (static pose) T ∈ SO(3) of the target robot refers to the position and orientation of the target robot relative to a certain reference point (such as the starting position or the world coordinate system) after it stops. The T therein represents the pose transformation matrix, belonging to the special orthogonal group SO(3), that is, the three-dimensional rotation group.

[0081] And obtaining the lidar data after stopping through the target radar (i.e., the static point cloud data) represents the instantaneous scanning result of the target robot on the surrounding environment in the static state, usually two-dimensional or three-dimensional point cloud data, including information such as obstacle distance and angle.

[0082] Then, it can be further determined whether the acquired static point cloud data is the first-frame point cloud data. If so, the static point cloud data can be used as the initial environmental point cloud data. If not, the relative pose (static pose) T∈SO(3) can be used as the initial prediction, and the lidar data (i.e., the static point cloud data) is transformed from the radar coordinate system to the world coordinate system. Then, the transformed lidar data (i.e., the static point cloud data) is registered with the environmental point cloud (such as the initial environmental point cloud data or the environmental point cloud data updated in the previous frame). The specific registration algorithm and implementation process are not limited and may include, but are not limited to, coordinate system transformation, initial pose initialization, and registration algorithm optimization and other processing steps to update the environmental point cloud data. And so on. After repeating the above steps multiple times until the coverage range of the updated environmental point cloud data can meet the preset conditions (such as the point cloud covers more than 70% of the entire environment), to ensure that the updated environmental point cloud data should cover the entire environment as much as possible.

[0083] S202: Use a spline curve to fit the updated environmental point cloud data to obtain the fitted environmental data; and perform sampling processing on the fitted environmental data to obtain the ideal environmental point cloud data.

[0084] In this embodiment, after obtaining the updated environmental point cloud data through step S201, in order to improve the evaluation accuracy of the motion compensation effect of the target radar installed on the target robot, a spline curve can be further used to fit the updated environmental point cloud data to obtain the fitted environmental data; and perform sampling processing on the fitted environmental data to obtain the ideal environmental point cloud data (which can be understood as the smooth real environmental point cloud data) for subsequent step S203.

[0085] Specifically, an optional implementation method is that after obtaining the updated environmental point cloud data, first, the discrete point cloud data in the updated environmental point cloud data can be spliced using a spline curve to complete the smoothness modeling of the environmental data and obtain the integrated smooth environmental data as the fitted environmental data. Then, N sampling points are uniformly selected from the fitted environmental data as the ideal environmental point cloud data, where the value of N is not limited and can be set as any positive integer greater than 1 according to the actual situation and empirical values, such as N can be set to 1000, etc.

[0086] Illustrate with an example: Take the actual environment as an artificially built approximately square environment with occlusions as an example, such as Figure 4As shown, in the spline curve fitting effect shown in the figure, the green part represents the schematic diagram of the spline curve (i.e., the smooth curve obtained by splicing the discrete point cloud data in the updated environmental point cloud data). The red part represents N sampling points evenly selected from the spline curve, constituting the ideal environmental point cloud data. The blue part represents the updated environmental point cloud data. It can be understood that Figure 4 The blue vacant part in it represents the occlusion part existing in the artificially built approximate square environment in the actual environment. After this application fills it up with a spline curve and samples it, the red part is obtained as the ideal environmental point cloud data, thereby improving the accuracy of the environmental point cloud.

[0087] In this way, through the form of interacting with the target robot and fitting with the spline curve in the above steps, it is possible to realize steps such as controlling the robot to rotate automatically, obtaining real-time point cloud, and splicing the point cloud, so as to obtain the representation of the ideal environmental point cloud data, such as Figure 3 As shown, as the evaluation basis, it can further improve the evaluation accuracy of the motion compensation effect of the target radar in the subsequent steps.

[0088] S203: Calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data.

[0089] It should be noted that during the process of controlling the free movement of the target robot and performing motion compensation on the target radar installed in the target robot, as time goes by, the distortion of the points obtained later is usually greater. And a small number of points with distortion can often better evaluate the de-distortion (i.e., motion compensation) effect. Therefore, this application proposes a distance penalty function and an evaluation optimization method based on this function. The main idea of the evaluation is that the farther away from the ideal environmental point cloud data, the stronger the penalty, and the more the result should be reflected.

[0090] Based on this, in this embodiment, after obtaining the ideal environmental point cloud data through step S202, the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data can be further calculated and represented by score for use in the subsequent step S204.

[0091] Specifically, an optional implementation is that after obtaining the ideal environmental point cloud data, further, an existing or future Euclidean distance calculation method can be used to calculate the Euclidean distance between each point in the real-time point cloud data and each point in the ideal environmental point cloud data, and based on the obtained Euclidean distance calculation result, determine the point in the real-time point cloud data that is closest to the point in the ideal environmental point cloud data, that is, the point with the smallest Euclidean distance. Then, an exponential function and a preset distance threshold (the specific value is not limited and can be set according to the actual situation and empirical values) are used to calculate the distance score between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it.

[0092] In this implementation, since the exponential function is highly sensitive to numerical values, therefore, the present application constructs a distance penalty function based on the exponential function, and the specific calculation formula is as follows:

[0093] score+=exp((dis-dis max ) / dis max )-exp(-1.0)

[0094] s=score / num

[0095] Where score represents the distance score between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it; score+ represents the cumulative value of the distance scores between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it; dis represents the Euclidean distance between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it; dis max represents the preset distance threshold, the specific value is not limited and can be set according to the actual situation and empirical values, for example, its value can be set to 0.5m, etc.; s represents the average value of the distance scores between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it.

[0096] It can be understood that when the Euclidean distance between the point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it is large, it can be clearly reflected in the score value score.

[0097] For example: when dis max takes the value of 0.5m, the function distribution diagram of the score value score can be as Figure 5 shown. When the Euclidean distance between the point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it reaches 0.75, the score value score increases significantly, and when the distance reaches 1.0m, it shows a rapid increase.

[0098] S204: Use the distance score to evaluate the motion compensation effect of the target radar and obtain an evaluation result.

[0099] In this embodiment, after calculating the distance score score between each point in the real-time point cloud data and the point with the closest distance in the ideal environment point cloud data through step S203, the distance score score can be further used to evaluate the motion compensation effect of the target radar, and an evaluation result can be obtained. Specifically, the larger the distance score score is, it indicates that the distance between the corresponding point in the real-time point cloud data and the point with the closest distance in the ideal environment point cloud data is larger, and this point is more likely to be an outlier. Therefore, the value of the distance score score can be used to find more outliers caused by motion compensation. The more outliers there are, the worse the evaluation result of the motion compensation effect of the target radar will be. On the contrary, the fewer outliers there are, the better the evaluation result of the motion compensation effect of the target radar will be.

[0100] Illustrative example: Based on the above example, when dis max takes a value of 0.5 m, the function distribution diagram of the score value score can be as Figure 5 shown. Then, it can be seen through Figure 5 that when the Euclidean distance between the point in the real-time point cloud data and the point with the closest distance in the ideal environment point cloud data reaches 0.75, the score value score increases significantly, and when the distance reaches 1.0 m, it shows a rapid increase, which can also quickly reflect the value of the score value score, so as to find more outliers caused by motion compensation, and indirectly reflects the motion compensation effect (that is, the more outliers there are, the worse the compensation effect; the fewer outliers there are, the better the compensation effect). Compared with the existing calculation methods of ICP and other score thresholds, it is more sensitive to numerical values, and the obtained evaluation result is more accurate.

[0101] In addition, in order to further verify the effectiveness of the radar motion compensation effect evaluation provided by this application, an optional implementation method is that during the process of controlling the target robot to move freely, when using the odometer (Odometry, odom) composed of an Inertial Measurement Unit (IMU) and a wheel speedometer to perform motion compensation on the target radar installed in the target robot, the evaluation result can be verified by adding position noise and / or attitude noise to the odometer (odom), and a verification result can be obtained. That is to say, by adding position noise and / or attitude noise to the odometer (odom), it can be verified at what noise value the radar motion compensation effect evaluation method provided by this application can work.

[0102] Specifically, in this implementation, if position noise is added to the odometer and the score threshold exceeding 0.1 is set as a problem with motion compensation, then when the preset distance threshold dis in the above distance penalty function calculation formula max takes a value of 0.015 m, taking the actual environment as an approximately square environment built artificially as an example, the standard deviation unit of the Gaussian distribution of the noise is meters (m), and the [mean, variance] of the Gaussian distribution of the noise and the related values of the score value score can be as shown in Table 1 below:

[0103] Gaussian Distribution [0，0] [0，0.025] [0，0.05] [0，0.075] score mean (i.e., s) 0.008 0.046 0.1956 0.9515 score standard deviation 0.0043 0.0442 0.1218 1.188 score maximum difference 0.028 0.321126 0.6925 5.7269

[0104] Table 1

[0105] It can be seen from Table 1 that when the variance of the noise gradually increases, the score mean (i.e., s), standard deviation, and maximum value change significantly, so that the size of the distortion can be judged numerically.

[0106] Moreover, the above data can also be presented as a visualization diagram as Figure 6 shown, that is, some results after adding Gaussian error to the position are shown as Figure 6 . In Figure 6 , the blue part represents the environmental point cloud data obtained by two-frame registration. For example, the target robot can move once from rest and then stop again to achieve two-frame registration to obtain the environmental point cloud data. The specific registration method will not be elaborated here. The red part represents the point cloud data after motion compensation. The score value in this figure is 0.321. In this way, it can be clearly seen from Figure 6 that the difference between the point cloud after motion compensation and the environmental point cloud is large, that is, the red part and the blue part do not completely overlap, and there are multiple breakpoints (i.e., missing parts) in the red part relative to the blue part. At this time, the appropriate score threshold can be set to screen the point cloud with problems in motion compensation.

[0107] Similarly, if attitude noise is added to the odometer and the score threshold exceeding 0.1 is still set as a problem with motion compensation, then when the preset distance threshold dis in the above distance penalty function calculation formula max takes a value of 0.015 m, still taking the actual environment as an approximately square environment built artificially as an example, the standard deviation unit of the Gaussian distribution of the noise is degrees (°), and the [mean, variance] of the Gaussian distribution of the noise and the related values of the score value score can be as shown in Table 2 below:

[0108] Gaussian Distribution [0，1] [0，3] [0，5] score mean (i.e., s) 0.0085 0.0229 0.0681 score standard deviation 0.0047 0.0148 0.0631 score maximum difference 0.0234 0.0694 0.41091

[0109] Table 2

[0110] As can be seen from Table 2, when the variance of the noise gradually increases, the mean value of the score (i.e., s), the standard deviation, and the maximum value change significantly, so that the size of the distortion can be judged numerically.

[0111] Moreover, the above data can also be displayed as a Figure 7 visualization diagram as shown, that is, part of the results after adding Gaussian error to the pose are shown as Figure 7 . In Figure 7 , the blue part still represents the environmental point cloud data obtained through two-frame registration. For example, the target robot can move from rest once and then stop again to achieve two-frame registration to obtain the environmental point cloud data. The specific registration method will not be elaborated here. The red part represents the point cloud data after motion compensation. The score value in this figure is 0.129. In this way, it can be clearly seen from Figure 7 that there is a large gap between the point cloud after motion compensation and the environmental point cloud, that is, the red part and the blue part do not completely overlap, and there are multiple breakpoints (i.e., missing parts) in the red part relative to the blue part. At this time, the point cloud with problems in motion compensation can be screened by setting an appropriate score threshold.

[0112] In summary, a method for evaluating the radar motion compensation effect provided in this embodiment is as follows: First, during the process of controlling the target robot to move freely, motion compensation is performed on the target radar installed in the target robot, and real-time point cloud data is obtained through the target radar to update the environmental point cloud data using the real-time point cloud data. Then, the spline curve is used to fit the environmental point cloud data to obtain the fitted environmental data; and the fitted environmental data is sampled to obtain the ideal environmental point cloud data; Next, the distance score between each point in the real-time point cloud data and the point closest to it in the ideal environmental point cloud data is calculated; Furthermore, the motion compensation effect of the target radar can be evaluated using the distance score to obtain the evaluation result.

[0113] It can be seen that when evaluating the motion compensation effect of the target radar in this application, the real-time point cloud data obtained by the target radar during the free movement of the target robot is first fitted using the spline curve to obtain smooth fitted environmental data, and then the ideal environmental point cloud data is sampled from it. Furthermore, the distance score between each point in the real-time point cloud data and the point closest to it in the ideal environmental point cloud data can be calculated to more accurately discover the outliers caused by motion compensation, so as to effectively evaluate the motion compensation effect of the target radar. Compared with the prior art in which the environmental data collected in the first frame is used as the relative environmental true value or the motion compensation effect is verified based on the matching score of the Euclidean distance (it is difficult to discover outliers), the accuracy of the evaluation result is higher.

[0114] Second Embodiment

[0115] This embodiment will introduce an evaluation device for the radar motion compensation effect. For related content, please refer to the above method embodiment.

[0116] See Figure 8 , which is a schematic diagram of the composition of an evaluation device for the radar motion compensation effect provided in this embodiment. The device 800 includes:

[0117] An acquisition unit 801, configured to perform motion compensation on the target radar installed in the target robot during the process of controlling the target robot to move freely, and acquire real-time point cloud data through the target radar, so as to update the environmental point cloud data by using the real-time point cloud data;

[0118] A fitting unit 802, configured to fit the environmental point cloud data by using a spline curve to obtain fitted environmental data; and perform sampling processing on the fitted environmental data to obtain ideal environmental point cloud data;

[0119] A calculation unit 803, configured to calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data;

[0120] An evaluation unit 804, configured to evaluate the motion compensation effect of the target radar by using the distance score to obtain an evaluation result.

[0121] In an implementation manner of this embodiment, the acquisition unit 801 includes:

[0122] An acquisition subunit, configured to acquire the static pose of the target robot after free movement, and after the target robot stops, acquire static point cloud data through the target radar;

[0123] A judgment subunit, configured to judge whether the static point cloud data is the first-frame point cloud data. If so, use the static point cloud data as the initial environmental point cloud data; if not, register the static point cloud data with the existing environmental point cloud data according to the static pose to update the environmental point cloud data; and so on, until the coverage range of the environmental point cloud data meets the condition, and obtain the updated environmental point cloud data.

[0124] In an implementation manner of this embodiment, the fitting unit 802 is specifically configured to:

[0125] Use a spline curve to splice the discrete point cloud data in the updated environmental point cloud data, complete the smoothness modeling of the environmental data, and obtain the integrated smooth environmental data as the fitted environmental data.

[0126] In an implementation manner of this embodiment, the fitting unit 802 is specifically configured to:

[0127] Uniformly select N sampling points from the fitted environmental data as the ideal environmental point cloud data; N is a positive integer greater than 1.

[0128] In one implementation of this embodiment, the calculation unit 803 includes:

[0129] A first calculation subunit, configured to calculate the Euclidean distance between each point in the real-time point cloud data and each point in the ideal environmental point cloud data, and determine, according to the obtained Euclidean distance calculation result, the point in the real-time point cloud data that is closest to the point in the ideal environmental point cloud data;

[0130] A second calculation subunit, configured to calculate the distance score between each point in the real-time point cloud data and the point in the ideal environmental point cloud data that is closest to it by using an exponential function and a preset distance threshold.

[0131] In one implementation of this embodiment, the obtaining unit 801 is specifically configured to:

[0132] During the process of controlling the target robot to move freely, use an odometer composed of an inertial measurement unit IMU and a wheel speedometer to perform motion compensation on the target radar installed in the target robot.

[0133] In one implementation of this embodiment, the device further includes:

[0134] A verification unit, configured to verify whether the evaluation result is valid by adding position noise and / or attitude noise to the odometer, and obtain a verification result.

[0135] Furthermore, an embodiment of the present application further provides a radar motion compensation effect evaluation device, including: a processor, a memory, and a system bus;

[0136] The processor and the memory are connected through the system bus;

[0137] The memory is used to store one or more programs, and the one or more programs include instructions, and when the instructions are executed by the processor, the processor is caused to execute any implementation method of the above-mentioned radar motion compensation effect evaluation method.

[0138] Furthermore, an embodiment of the present application further provides a computer-readable storage medium, in which instructions are stored, and when the instructions are run on a terminal device, the terminal device is caused to execute any implementation method of the above-mentioned radar motion compensation effect evaluation method.

[0139] Furthermore, an embodiment of the present application also provides a computer program product. When the computer program product runs on a terminal device, it causes the terminal device to execute any one of the implementation methods of the above-mentioned radar motion compensation effect evaluation method.

[0140] From the description of the above embodiments, those skilled in the art can clearly understand that all or part of the steps in the above method embodiments can be implemented by means of software plus a necessary general hardware platform. Based on this understanding, the technical solution of the present application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. The computer software product can be stored in a storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network communication device such as a media gateway, etc.) to execute the methods described in various embodiments or some parts of the embodiments of the present application.

[0141] It should be noted that the embodiments in this specification are described in a progressive manner. The key point of each embodiment is to illustrate the differences from other embodiments. The same or similar parts among the embodiments can be referred to each other. For the devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0142] It should also be noted that in this article, relational terms such as first and second are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements not only includes those elements, but also includes other elements not expressly listed, or elements inherent to such process, method, article or device. Without further limitation, an element defined by the statement "including a..." does not exclude the existence of additional identical elements in the process, method, article or device including the element.

[0143] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for evaluating the radar motion compensation effect, characterized in that, Including: During the process of controlling the free movement of the target robot, perform motion compensation on the target radar installed in the target robot, and obtain real-time point cloud data through the target radar, so as to update the environmental point cloud data by using the real-time point cloud data; Use a spline curve to fit the updated environmental point cloud data to obtain fitted environmental data; and perform sampling processing on the fitted environmental data to obtain ideal environmental point cloud data; Calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data; Use the distance score to evaluate the motion compensation effect of the target radar to obtain an evaluation result.

2. The method according to claim 1, characterized in that, The obtaining real-time point cloud data through the target radar to update the environmental point cloud data by using the real-time point cloud data includes: Obtain the static pose of the target robot after free movement, and after the target robot stops, obtain static point cloud data through the target radar; Judge whether the static point cloud data is the first-frame point cloud data. If so, use the static point cloud data as the initial environmental point cloud data; if not, register the static point cloud data with the existing environmental point cloud data according to the static pose to update the environmental point cloud data; and so on until the coverage range of the environmental point cloud data meets the conditions to obtain the updated environmental point cloud data.

3. The method according to claim 1, wherein The using a spline curve to fit the updated environmental point cloud data to obtain fitted environmental data includes: Use a spline curve to splice the discrete point cloud data in the updated environmental point cloud data, complete the smoothness modeling of the environmental data, and obtain the integrated smooth environmental data as the fitted environmental data.

4. The method according to claim 1, wherein The performing sampling processing on the fitted environmental data to obtain ideal environmental point cloud data includes: Uniformly select N sampling points from the fitted environmental data as the ideal environmental point cloud data; N is a positive integer greater than 1.

5. The method according to claim 1, wherein The calculating the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data includes: Calculate the Euclidean distance between each point in the real-time point cloud data and each point in the ideal environmental point cloud data, and determine the point with the closest distance between each point in the real-time point cloud data and the ideal environmental point cloud data according to the obtained Euclidean distance calculation result; Use an exponential function and a preset distance threshold to calculate the distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data.

6. The method according to any one of claims 1-5, characterized in that, The performing motion compensation on the target radar installed in the target robot during the process of controlling the free movement of the target robot includes: During the process of controlling the free movement of the target robot, use an odometer composed of an inertial measurement unit (IMU) and a wheel speedometer to perform motion compensation on the target radar installed in the target robot.

7. The method according to claim 6, wherein After the using the distance score to evaluate the motion compensation effect of the target radar to obtain an evaluation result, the method further includes: Verify whether the evaluation result is valid by adding position noise and / or attitude noise to the odometer to obtain a verification result.

8. A device for evaluating the radar motion compensation effect, characterized in that, Including: An acquisition unit, configured to perform motion compensation on a target radar installed in the target robot during the process of controlling the target robot to move freely, and acquire real-time point cloud data through the target radar, so as to update environmental point cloud data by using the real-time point cloud data; A fitting unit, configured to fit the environmental point cloud data by using a spline curve to obtain fitted environmental data; and perform sampling processing on the fitted environmental data to obtain ideal environmental point cloud data; A calculation unit, configured to calculate a distance score between each point in the real-time point cloud data and the point with the closest distance in the ideal environmental point cloud data; An evaluation unit, configured to evaluate the motion compensation effect of the target radar by using the distance score to obtain an evaluation result.

9. A radar motion compensation effect evaluation device, characterized in that Including: A processor, a memory, and a system bus; The processor and the memory are connected through the system bus; The memory is configured to store one or more programs, and the one or more programs include instructions, and when the instructions are executed by the processor, the processor is caused to execute the method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, Instructions are stored in the computer-readable storage medium, and when the instructions are run on a terminal device, the terminal device is caused to execute the method according to any one of claims 1-7.

Images