Speed prediction method, device, equipment, storage medium and autonomous vehicle

Abstract

The disclosure provides a speed prediction method, device, equipment, storage medium and an autonomous vehicle, relates to the technical field of data processing, and particularly relates to the fields of artificial intelligence, autonomous driving and computer vision. A specific implementation scheme is as follows: signal source data of a target object and a plurality of candidate prediction speeds are acquired, the signal source data comprising original data of at least one sensor; the confidence of each candidate prediction speed is calculated by using the original data of the at least one sensor; and the target prediction speed of the target object is determined from the plurality of candidate prediction speeds according to the confidence of each candidate prediction speed. According to the technology of the disclosure, the accuracy of the speed prediction of the target object is improved, and the technology has a wide range of applications.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to the fields of artificial intelligence, autonomous driving, and computer vision technology, specifically to speed prediction methods, devices, equipment, storage media, and autonomous vehicles. Background Technology

[0002] In related technologies, speed prediction of obstacles in autonomous driving scenarios typically involves fusing predicted speed values ​​for obstacle levels from different sensors to obtain a fused speed value for the obstacle. However, different sensors exhibit varying prediction accuracy in different scenarios. For example, camera sensors have low sensitivity in low-speed scenarios, radar sensors have low accuracy in predicting lateral speeds, and lidar has low sensitivity when switching between static and dynamic states. Therefore, speed fusion schemes in these technologies cannot guarantee high fusion accuracy for different scenarios. Summary of the Invention

[0003] This disclosure provides a speed prediction method, apparatus, device, storage medium, and autonomous vehicle.

[0004] According to one aspect of this disclosure, a speed prediction method is provided, comprising:

[0005] Acquire signal source data of the target object and multiple candidate prediction velocities, wherein the signal source data includes raw data from at least one sensor;

[0006] Calculate the confidence level for each candidate predicted velocity using raw data from at least one sensor;

[0007] The target predicted velocity of the target object is determined from multiple candidate predicted velocities based on the confidence level of each candidate predicted velocity.

[0008] According to another aspect of this disclosure, a speed prediction device is provided, comprising:

[0009] The acquisition module is used to acquire signal source data of the target object and multiple candidate prediction velocities. The signal source data includes raw data from at least one sensor.

[0010] A confidence calculation module is used to calculate the confidence level of each candidate prediction velocity using raw data from at least one sensor.

[0011] The target prediction speed determination module is used to determine the target prediction speed of the target object from multiple candidate prediction speeds based on the confidence level of each candidate prediction speed.

[0012] According to another aspect of this disclosure, an electronic device is provided, comprising:

[0013] At least one processor; and

[0014] The memory is communicatively connected to the at least one processor; wherein,

[0015] The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods in any embodiment of this disclosure.

[0016] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions for causing a computer to perform the methods of any embodiment of this disclosure.

[0017] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods of any embodiment of this disclosure.

[0018] According to another aspect of this disclosure, an autonomous vehicle is provided, including a speed prediction device and / or an electronic device according to any embodiment of this disclosure.

[0019] The technology disclosed herein improves the accuracy of speed prediction and has a wide range of applications, enabling accurate prediction of the speed of target objects in various scenarios.

[0020] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0021] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein:

[0022] Figure 1 A flowchart illustrating a speed prediction method according to an embodiment of the present disclosure is shown;

[0023] Figure 2 A detailed flowchart is shown illustrating the calculation of the confidence level of each candidate predicted velocity according to an embodiment of the present disclosure.

[0024] Figure 3 A detailed flowchart illustrating the velocity prediction method according to an embodiment of the present disclosure for determining the translational relationship between two point cloud datasets is shown.

[0025] Figure 4 A detailed flowchart is shown for determining the overlap between two point cloud datasets using a velocity prediction method according to an embodiment of this disclosure.

[0026] Figure 5A detailed flowchart illustrating the speed prediction method according to an embodiment of the present disclosure for obtaining multiple candidate prediction speeds is shown.

[0027] Figure 6 A detailed flowchart is shown as an example of obtaining multiple candidate prediction velocities using a velocity prediction method according to an embodiment of the present disclosure;

[0028] Figure 7 A detailed flowchart is shown, illustrating yet another example of obtaining multiple candidate predicted velocities using a velocity prediction method according to an embodiment of the present disclosure;

[0029] Figure 8 A detailed flowchart is shown, illustrating yet another example of obtaining multiple candidate predicted velocities using a velocity prediction method according to an embodiment of the present disclosure;

[0030] Figure 9 A detailed flowchart is shown, illustrating yet another example of obtaining multiple candidate predicted velocities using a velocity prediction method according to an embodiment of the present disclosure;

[0031] Figure 10 The diagram illustrates an application example of the speed prediction method according to an embodiment of the present disclosure.

[0032] Figure 11 The diagram illustrates a specific application example of the speed prediction method according to an embodiment of the present disclosure, which determines the confidence level of each candidate predicted speed.

[0033] Figure 12 A block diagram of a speed prediction apparatus according to an embodiment of the present disclosure is shown;

[0034] Figure 13 This is a block diagram of an electronic device used to implement the speed prediction method of the embodiments of this disclosure. Detailed Implementation

[0035] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0036] The following is for reference. Figures 1 to 11 A speed prediction method according to embodiments of the present disclosure is described.

[0037] like Figure 1 As shown, the speed prediction method according to an embodiment of this disclosure includes the following steps:

[0038] S101: Acquire signal source data of the target object and multiple candidate prediction velocities, wherein the signal source data includes raw data from at least one sensor;

[0039] S102: Using raw data from at least one sensor, calculate the confidence level for each candidate predicted velocity;

[0040] S103: Determine the target prediction speed of the target object from multiple candidate prediction speeds based on the confidence level of each candidate prediction speed.

[0041] Depending on the application scenario, the target object can be different speed observation objects. For example, in the application scenario of autonomous driving, the target object is an obstacle that may interfere with the normal driving of autonomous vehicles, specifically autonomous vehicles, pedestrians, etc.

[0042] In the following description of embodiments of this disclosure, the speed prediction method is applied to an autonomous driving scenario and the target object is an obstacle.

[0043] For example, signal source data refers to data observed on a target object, used to characterize the target object's position, shape, and motion information.

[0044] In step S101, the raw sensor data included in the signal source data refers to the unprocessed and uncompressed raw data generated by the sensor after data acquisition. In this embodiment of the disclosure, the sensor may include at least one of a LiDAR sensor, a Radar sensor, and a camera sensor, and the signal source data may specifically include at least one of point cloud data output by the LiDAR sensor, point cloud data output by the Radar sensor, and RGB data output by the camera sensor.

[0045] The multiple candidate predicted speeds corresponding to the target object can include multiple candidate predicted speeds obtained by processing the signal source data accordingly. For example, the raw data from multiple sensors can be processed separately to obtain candidate predicted speeds corresponding to different sensors. It can also include candidate predicted speeds of the target object directly acquired. For example, a sensor can directly output the obstacle-level speed of the target object based on the acquired raw data; or, for example, the roadside perceived speed of the target object sent by roadside sensing devices can be directly received via V2X (vehicle-to-X wireless communication technology) as a candidate predicted speed.

[0046] For example, in step S102, point cloud data from a lidar sensor or a radar sensor can be fused with each candidate prediction speed to calculate the confidence level corresponding to each candidate prediction speed. It is understood that a higher confidence level for a candidate prediction speed indicates higher reliability or accuracy; conversely, a lower confidence level indicates lower reliability or accuracy.

[0047] In a specific example, the translation relationship of the point cloud data corresponding to the candidate prediction velocity at two different reference times can be calculated based on the reference velocity values ​​at those two different reference times. The two point cloud data sets are then aligned according to this translation relationship. The confidence level of the candidate prediction velocity can be obtained based on the degree of overlap between the aligned point cloud data sets. A higher degree of overlap indicates a higher confidence level for the candidate prediction velocity; conversely, a lower degree of overlap indicates a lower confidence level.

[0048] For example, in step S103, for each of the calculated candidate prediction speeds, the candidate prediction speed with the highest confidence is selected as the target prediction speed corresponding to the target object.

[0049] The following is a specific example illustrating the application of the speed prediction method according to embodiments of this disclosure in an autonomous driving scenario.

[0050] In a specific example, the speed prediction method of this disclosure can be used for an autonomous vehicle equipped with multiple sensors, including a lidar sensor, a millimeter-wave radar sensor, and a camera sensor. Specifically, it receives signal source data about a target obstacle collected by the multiple sensors, namely, first point cloud data output by the lidar sensor, second point cloud data output by the millimeter-wave radar sensor, and RGB data output by the camera sensor. The target obstacle can be an autonomous vehicle or a pedestrian.

[0051] Multiple candidate predicted velocities for the target obstacle are obtained. These multiple candidate predicted velocities can be obtained by processing the raw data from various sensors separately.

[0052] For example, for the first point cloud data output by a LiDAR sensor, after corresponding processing and calculation, a first preprocessing prediction speed and a first post-processing prediction speed at the obstacle level are obtained. The first post-processing prediction speed is obtained by applying filtering, smoothing, and verification to the first preprocessing prediction speed. For the second point cloud data output by a millimeter-wave radar, after corresponding processing and calculation, a second preprocessing prediction speed and a second post-processing prediction speed at the obstacle level are obtained. The second post-processing prediction speed is obtained by applying filtering, smoothing, and verification to the second preprocessing prediction speed. For the RGB data output by a camera sensor, after corresponding processing and calculation, a third preprocessing prediction speed and a cross-sectional prediction speed at the obstacle level are obtained. The cross-sectional prediction speed is obtained by horizontally splitting the third preprocessing prediction speed.

[0053] For example, the first post-processing prediction speed, the second post-processing prediction speed, and the cross-cutting prediction speed can be directly received from multiple sensors after processing and calculating their respective raw data.

[0054] Furthermore, the candidate predicted speed can be obtained through other methods. For example, at least two of the aforementioned first post-processed predicted speed, second post-processed predicted speed, and cross-sectional predicted speed can be concatenated to obtain a concatenated speed as a candidate predicted speed. Alternatively, the roadside sensing speed of the target obstacle can be directly received from roadside sensing devices via V2X and used as a candidate predicted speed. Yet another example is the compensation speed of the target obstacle, calculated based on its historical motion characteristics, which can also be used as a candidate predicted speed.

[0055] For the multiple candidate predicted velocities of the target obstacle, each candidate predicted velocity is fused with the first point cloud data output by the lidar sensor. The confidence level of each candidate predicted velocity is determined based on the fusion result. Finally, based on the confidence level, the candidate predicted velocity with the highest confidence level is selected as the target predicted velocity of the target obstacle.

[0056] It should be noted that, compared to the speed prediction methods for target objects in related technologies, directly fusing the obstacle-level predicted speeds output by each sensor results in poor accuracy of the fused predicted speeds because the accuracy of the predicted speeds output by each sensor varies in different scenarios. In other words, the weighting coefficients of the predicted speeds output by each sensor cannot be unified for different scenarios. As a result, the accuracy of the fused predicted speed is poor and it cannot be applied to any scenario.

[0057] The speed prediction method according to the embodiments of this disclosure integrates multiple candidate prediction speeds with the raw data of the sensor to obtain the confidence level corresponding to each candidate prediction speed. The target prediction speed determined based on the confidence level is more accurate, especially for speed prediction in low-speed motion scenarios of the target object. It effectively reduces the probability of false alarms for stationary target objects and can accurately predict the speed of the target object in various scenarios, thus having a wide range of applications.

[0058] like Figure 2 As shown, in one embodiment, the sensor's raw data includes point cloud data at at least two different reference times, and step S102 specifically includes the following steps:

[0059] S201: For each candidate prediction velocity, determine the reference velocity value corresponding to the candidate prediction velocity at two different reference times;

[0060] S202: Determine the translation relationship between two point cloud data based on two reference velocity values;

[0061] S203: Determine the overlap between two point cloud datasets based on the translation relationship;

[0062] S204: Determine the confidence level of the candidate prediction velocity based on the degree of overlap.

[0063] For example, the point cloud data can be dense point cloud data output by a lidar sensor or sparse point cloud data output by a millimeter-wave radar sensor. Preferably, in this embodiment, the point cloud data can be dense point cloud data output by a lidar sensor to improve the accuracy of the confidence calculation for the selected prediction velocity. It is understood that the raw data from the sensor can specifically be multi-frame point cloud data output by the lidar sensor, i.e., point cloud data corresponding to multiple time points. Specifically, the point cloud data at two different reference time points can be any two adjacent frames of point cloud data.

[0064] For example, based on the reference velocity values ​​corresponding to the candidate prediction velocity at two different reference times, the relative displacement relationship of the target object at the two different reference times can be estimated. Based on the relative displacement relationship, the translation relationship of the point cloud data at the two different reference times can then be determined. The two point cloud data are then translated and aligned according to the translation relationship, and the degree of overlap between the two point cloud data is determined based on the alignment result.

[0065] Understandably, a higher degree of overlap indicates a higher degree of matching between the candidate prediction speed and the point cloud data; conversely, a lower degree of overlap indicates a lower degree of matching between the candidate prediction speed and the point cloud data.

[0066] Therefore, the confidence level of each candidate prediction speed can be determined based on the degree of overlap between two point cloud data at different candidate prediction speeds.

[0067] Through the above implementation methods, high-precision point cloud data of the target object collected by lidar sensors or millimeter-wave radar sensors can be used to verify each candidate prediction speed. The confidence level of the final candidate prediction speed is relatively accurate, enabling the selection of the most accurate candidate prediction speed from multiple candidate speeds as the target prediction speed. Therefore, addressing the technical problem of varying accuracy among candidate prediction speeds in different scenarios, the above implementation methods can be used for verification, ensuring that the most accurate candidate prediction speed can be determined from multiple candidate prediction speeds in different scenarios.

[0068] like Figure 3 As shown, in one embodiment, step S202 specifically includes the following steps:

[0069] S301: Interpolate the two reference speed values ​​to obtain the average speed value of the two reference speed values;

[0070] S302: Determine the translation relationship between two point cloud data based on the average velocity value and the time difference between two different reference times.

[0071] For example, the interpolation process can be performed using any difference algorithm. This disclosure does not specifically limit this. For example, the nearest neighbor interpolation algorithm or the bilinear interpolation algorithm can be used, as long as the average velocity value of the two reference velocity values ​​can be calculated.

[0072] Understandably, by calculating the product of the average velocity value and the time difference between two different reference times, the displacement of the target object at the time difference between the two different reference times can be obtained, which is the translation relationship between the two point cloud data.

[0073] The above implementation method can quickly determine the translation relationship between two point cloud data at any candidate prediction speed, with a small computational load and relatively small computational resources required.

[0074] like Figure 4 As shown, in one embodiment, step S203 includes:

[0075] S401: Align the two point cloud data according to the translation relationship to obtain the alignment result of the two point cloud data. The alignment result includes multiple corresponding point groups of the two point cloud data.

[0076] S402: Calculate the logarithmic value of the Mahalanobis distance for each of multiple corresponding point groups;

[0077] S403: Determine the overlap between two point cloud data based on the logarithmic value of the Mahalanobis distance between each of the multiple corresponding point groups.

[0078] For example, the overlap C of two point cloud datasets can be calculated using the following formula:

[0079]

[0080] Used to represent the three-dimensional coordinate vector and RGB vector of the first point in the first point cloud data. The three-dimensional coordinate vector and RGB vector of the second point in the second point cloud data are used to represent the second point. `mahanalobis()` is used to represent the Mahalanobis distance, where the first and second points correspond to each other. The logarithmic function is L(x) = lg(x + 0.001) - lg(0.001), n i Used to represent the number of points contained in the first point cloud data and the second point cloud data.

[0081] The first point cloud data corresponds to the point cloud data at the first reference time. The second point cloud data corresponds to the point cloud data at the second reference time, which is then translated according to a translation relationship. In other words, the second point cloud data is obtained by translating the point cloud data corresponding to the second reference time based on the first point cloud data. The number of points n contained in the first and second point cloud data is... i That is, the number of corresponding point groups.

[0082] It is understandable that for any first point in the first cloud data, there is a second point in the second cloud data that is closest to the first point, and the corresponding first and second points form a corresponding point group.

[0083] For any corresponding point set, calculate the Mahalanobis distance between the first and second points based on the 3D coordinate vectors and RGB vectors of the first and second points, and the 3D coordinate vectors and RGB vectors of the second point. Then, use the logarithmic function L(x) = lg(x+0.001) - lg(0.001) to differentiate the Mahalanobis distance between the first and second points, obtaining the derivative of the Mahalanobis distance for that corresponding point set. Average the derivatives of the Mahalanobis distances for all corresponding point sets to obtain the overlap between the first and second point cloud data.

[0084] Through the above implementation method, the point cloud data corresponding to two different reference times are aligned according to the translation relationship corresponding to the candidate prediction velocity, and the overlap between the two point cloud data is calculated based on the alignment result, thereby obtaining the confidence level of the candidate prediction velocity.

[0085] In one implementation, the sensor's raw data includes point cloud data at at least two different reference times, and step S101 further includes:

[0086] Acquire point cloud data to be processed at at least two distinct reference times;

[0087] The point cloud data to be processed is preprocessed to obtain point cloud data at at least two different reference times.

[0088] Preprocessing of the point cloud data to be processed may specifically include at least one of downsampling, floating-point precision reduction, and noise filtering.

[0089] Therefore, by preprocessing the point cloud data to be processed, the complexity of the obtained point cloud data can be reduced, thereby reducing the amount of computation in the confidence calculation process, improving computational efficiency and reducing the occupation of computing resources.

[0090] like Figure 5 As shown, in one embodiment, step S101 includes:

[0091] S501: Based on the point cloud data from the lidar sensor, determine the first preprocessing prediction speed and the first postprocessing prediction speed corresponding to the target object, wherein the postprocessing prediction speed is obtained by postprocessing based on the preprocessing prediction speed; and,

[0092] S502: Based on the point cloud data from the millimeter-wave radar sensor, determine the second preprocessing prediction speed and the second post-processing prediction speed corresponding to the target object, wherein the second post-processing prediction speed is obtained by post-processing based on the second preprocessing prediction speed; and...

[0093] S503: Based on the RGB data of the camera sensor, determine the third preprocessing prediction speed and the lateral splitting prediction speed corresponding to the target object. The lateral splitting prediction speed is obtained by lateral splitting based on the third preprocessing prediction speed.

[0094] S504: Determine at least two of the following: first preprocessing prediction speed, first postprocessing prediction speed, second preprocessing prediction speed, second postprocessing prediction speed, third preprocessing prediction speed, and lateral splitting prediction speed as at least two candidate prediction speeds.

[0095] For example, in step S501, the first preprocessing prediction speed corresponding to the target object is calculated based on the point cloud data of the lidar sensor, specifically including the following steps:

[0096] (1) Perform time synchronization and external parameter calibration on the input laser point cloud;

[0097] (2) Considering the sampling noise of lidar and the large amount of point cloud data, the point cloud data is preprocessed to reduce the amount of data and remove noise points;

[0098] (3) Obtain the point cloud information of the target object in the point cloud data, and segment the point cloud information of the target object;

[0099] (4) An unsupervised clustering algorithm is used to form point cloud clusters from the point cloud information of the target object;

[0100] (5) Feature extraction and classifiers are used to identify point cloud clusters to obtain the category information of the target object;

[0101] (6) Fit bounding boxes to point cloud clusters and calculate the attributes of point cloud clusters, such as center point, centroid, length, width and height.

[0102] (7) Output the velocity information of the point cloud clusters, i.e. the target object, as the first preprocessing prediction velocity.

[0103] After obtaining the first preprocessed prediction speed, a second postprocessed prediction speed with higher accuracy can be obtained by performing postprocessing such as filtering, smoothing, and verification on the first preprocessed prediction speed.

[0104] For example, in step S502, based on the point cloud data of the millimeter-wave radar sensor, the second preprocessing prediction speed and the second postprocessing prediction speed can be obtained in the same or similar manner as the example above.

[0105] For example, in step S503, any method known or known in the art to those skilled in the art can be used to process and calculate the RGB data acquired by the camera sensor to obtain the third preprocessing prediction speed. For instance, the depth data of the target object can be calculated using the RGB data output by the binocular camera sensors, and the third preprocessing prediction speed of the target object can be calculated based on the RGB data and the depth data.

[0106] After obtaining the third preprocessing prediction speed, the horizontal split prediction speed and the vertical split prediction speed are obtained by splitting the third preprocessing prediction speed horizontally and vertically.

[0107] Preferably, the first preprocessing prediction speed, the first postprocessing prediction speed, the second preprocessing prediction speed, the second postprocessing prediction speed, the third preprocessing prediction speed, and the horizontal splitting prediction speed can all be used as candidate prediction speeds to maximize the number of candidate prediction speeds available for selection.

[0108] Through the above implementation method, multiple candidate prediction speeds are obtained based on the point cloud data of the lidar sensor, the point cloud data of the millimeter-wave radar sensor, and the RGB data of the camera sensor, thereby increasing the number of candidate prediction speeds and expanding the selection range of target prediction speeds. This allows full utilization of the advantages of the raw data from different sensors in different scenarios, enabling more accurate target prediction speeds to be obtained in different scenarios.

[0109] like Figure 6 As shown, in one embodiment, step S101 further includes:

[0110] S601: Based on the first post-processing prediction speed, the second post-processing prediction speed, and the horizontal splitting prediction speed, at least two of them are spliced ​​together to obtain at least one splicing speed.

[0111] S602: Determine at least one stitching speed as at least one of a plurality of candidate prediction speeds.

[0112] For example, the first post-processing prediction speed and the second post-processing prediction speed can be separated and stitched together by projection to obtain the first stitched speed. It is understood that the velocity direction of the first post-processing prediction speed obtained from point cloud data of a LiDAR sensor is more accurate, while the velocity magnitude of the second post-processing prediction speed obtained from point cloud data of a millimeter-wave radar sensor is more accurate. By stitching together the first and second post-processing prediction speeds, the velocity direction and magnitude of the obtained first stitched speed are both more accurate.

[0113] Furthermore, the first post-processing prediction speed can be concatenated with the lateral splitting speed to obtain a second concatenated speed with more accurate lateral splitting speed and speed direction. And / or, the second post-processing prediction speed can be concatenated with the lateral splitting speed to obtain a third concatenated speed with more accurate lateral splitting speed and speed magnitude.

[0114] Through the above implementation method, by splicing at least two of the first post-processing prediction speed, the second post-processing prediction speed, and the horizontal splitting prediction speed to obtain at least one spliced ​​speed, the number of candidate prediction speeds can be further expanded, and the spliced ​​speed obtained after splicing is more accurate, which is more conducive to improving the accuracy of the target prediction speed.

[0115] like Figure 7 As shown, in one embodiment, step S101 further includes:

[0116] S701: Obtain the historical motion characteristics of the target object;

[0117] S702: Calculate the compensation velocity of the target object based on historical motion characteristics;

[0118] S703: The compensation speed of the target object is determined as one of several candidate prediction speeds.

[0119] For example, the historical motion characteristics of the target object may specifically include historical velocity information, historical acceleration information, or historical angular acceleration information of the target object. Based on the historical motion characteristics of the target object, the current compensation velocity of the target object can be calculated as the candidate predicted velocity of the target object.

[0120] Through the above implementation method, since there is a certain correlation between the historical motion characteristics of the target object and the current predicted speed of the target object, the obtained compensation speed can be observed as a candidate predicted speed, thereby further increasing the number of candidate predicted speeds.

[0121] like Figure 8 As shown, in one embodiment, the signal source data further includes roadside sensing data; step S101 further includes:

[0122] S801: Determine the roadside sensing speed of the target object based on roadside sensing data;

[0123] S802: The roadside perceived speed is determined as one of several candidate predicted speeds.

[0124] For example, roadside perception data can be collected by roadside sensors installed at the roadside. These roadside sensors may specifically include one or more of LiDAR sensors, radar sensors, and camera sensors. Furthermore, autonomous vehicles can receive roadside perception data from these sensors or directly receive roadside perception speeds derived from the roadside perception data via V2X technology.

[0125] Understandably, because the roadside sensors and the sensors on autonomous vehicles are positioned differently and have different fields of view, the roadside sensors can collect information outside the field of view of the sensors on autonomous vehicles, thus supplementing the blind spots of the raw data from the sensors on autonomous vehicles.

[0126] Therefore, the speed perceived from the roadside is more accurate in certain scenarios compared to the predicted speed based on the raw data from sensors on autonomous vehicles.

[0127] The above implementation method further increases the number of candidate predicted speeds, fills in the blind spots of sensors on autonomous vehicles, and thus expands the selection range of target candidate predicted speeds.

[0128] like Figure 9As shown, in one embodiment, step S101 includes:

[0129] S901: Based on the signal source data, determine multiple first predicted velocities corresponding to the target object; and,

[0130] S902: Based on at least one first prediction velocity, at least one second prediction velocity is obtained using an optimization algorithm;

[0131] S903: Determine at least one first prediction velocity and / or at least one second prediction velocity as at least one candidate prediction velocity among a plurality of candidate prediction velocities.

[0132] For example, in step S901, multiple first prediction speeds can be selected from the first preprocessing prediction speed, the first postprocessing prediction speed, the second preprocessing prediction speed, the second postprocessing prediction speed, the third preprocessing prediction speed, the lateral splitting prediction speed, the stitching speed, the compensation speed, and the roadside sensing speed obtained in the aforementioned steps.

[0133] In one example, the second prediction speed can be obtained by using an optimization algorithm based on a preset number of first prediction speeds randomly selected from a plurality of first prediction speeds.

[0134] In another example, for each first predicted speed, the confidence level of each first predicted speed can also be calculated using various implementations of step S102, and a preset number of first predicted speeds with the highest confidence levels can be selected from the multiple first predicted speeds as input to the optimization algorithm to obtain the second predicted speed.

[0135] It is understandable that the optimization algorithm may employ various search algorithms that are known to or will become known to those skilled in the art, in order to further optimize and obtain different second prediction speeds based on multiple first prediction speeds.

[0136] Preferably, in step S903, all first prediction velocities and all second prediction velocities can be used as candidate prediction velocities.

[0137] Through the above implementation method, after obtaining multiple first prediction velocities based on signal source data, a more refined second prediction velocity can be obtained by using an optimization algorithm. The accuracy of the obtained second prediction velocity is higher, thereby further expanding the selection range of candidate prediction velocities.

[0138] In one implementation, step S902 includes:

[0139] Based on at least one first predicted velocity, at least one second predicted velocity is obtained using a neighborhood search algorithm and / or a local minimum algorithm.

[0140] For example, the neighborhood search algorithm can specifically employ the Variable Neighborhood Search (VNS) algorithm. VNS is an improved local search algorithm where each iteration searches the neighborhood of the current solution. VNS continuously improves the neighborhood structure based on the neighborhood search algorithm, increasing the search range by changing the neighborhood. Generally, the larger the searched neighborhood, the better the quality of the optimal solution obtained from the local search, and therefore the higher the accuracy of the final solution, i.e., the second prediction speed.

[0141] For example, a local minimum algorithm can employ the Nelder-Mead algorithm. The Nelder-Mead algorithm is used to find the minimum of an objective function in a multidimensional space. It is a direct search method based on comparisons and is typically applied to nonlinear optimization problems where the derivative is unknown. For instance, an n-dimensional Nelder-Mead consists of n+1 test points (i.e., n test points randomly selected based on a first predicted value), forming a simplex. The objective function value is then calculated for each point, with the goal of finding a new test point to replace the old one, iterating the process. The simplest approach is to replace the worst point with the reflection point of the centroid (the mean of the first n points). If the reflection point is worse than the current point, the search can continue in the direction of the reflection point; if it is worse, all points are contracted towards a better direction.

[0142] The above implementation method can improve the optimization accuracy of the first prediction speed, thereby further improving the accuracy of the candidate prediction speed determined based on the second prediction speed.

[0143] The following reference Figure 10 and 11 An example application scenario of the speed prediction method according to embodiments of the present disclosure is described.

[0144] like Figure 10As shown, signal source data and multiple candidate predicted velocities are input into the observation candidate pool. The signal source data includes point cloud data from LiDAR sensors (LiDAR XYZ), millimeter-wave radar sensors (RadarXYZ), camera sensors (Camera RGB), and roadside perception data (V2X attributes). The multiple candidate predicted velocities include a first preprocessed predicted velocity (Raw_speed) and a first post-processed predicted velocity (Speed) obtained from the radar sensor point cloud data; a second preprocessed predicted velocity (Raw_speed) and a second post-processed predicted velocity (Speed) obtained from the millimeter-wave radar sensor point cloud data; a third preprocessed predicted velocity (Speed) and a lateral splitting velocity obtained from the camera sensor RGB data; a historical / compensated velocity obtained based on the historical motion features of the target object; a roadside perception velocity obtained from the roadside perception data; and a stitched velocity (LR velocity obtained by stitching the first and second post-processed predicted velocities, and RC velocity obtained by stitching the second post-processed predicted velocity and the lateral splitting velocity).

[0145] Based on the candidate prediction speeds mentioned above, an optimization search is performed using the neighborhood search algorithm and the Nelder-Mead algorithm to obtain an optimized prediction speed, which is then added to the observation candidate pool as a candidate prediction speed.

[0146] For each candidate predicted velocity, point cloud data (i.e., high-dimensional point cloud information) is used to evaluate each candidate predicted velocity, and the confidence level of each candidate predicted velocity is obtained. In addition, the evaluation can also be based on the matching degree between the velocity and displacement of each candidate predicted velocity and the point cloud data, as well as on the occlusion and collision risks.

[0147] Based on confidence level and / or other evaluation results, a target predicted velocity is selected from multiple candidate predicted velocities and reported. The reported target predicted velocity is filtered, and outlier data is removed to prevent false alarms.

[0148] like Figure 11 As shown, the confidence level of each candidate prediction speed is obtained by evaluating each candidate prediction speed using point cloud data, specifically including the following steps:

[0149] (1) Each candidate predicted velocity is divided into horizontal and vertical components in a horizontal coordinate system to show the direction and magnitude of each candidate predicted velocity. Among them, the candidate predicted velocities include the obstacle level velocity output by each sensor, the stitching velocity, and the optimized predicted velocity obtained by the optimization algorithm.

[0150] (2) For each candidate prediction speed at t k-1 and tk The velocity values ​​and directions at two different reference times are interpolated to obtain the average velocity of the candidate predicted velocity. Based on this average velocity, the target object's velocity at time t is calculated. k-1 and t k Displacement at two different reference times, t is determined based on displacement. k-1 and t k The translation relationship between point cloud data A and B at two different reference times. Alignment processing of point cloud data A and B is then performed based on the translation relationship. Figure 11 The results show the point cloud alignment at different candidate prediction velocities, such as the point cloud alignment at obstacle-level velocity (C velocity) output by the camera sensor, the point cloud alignment at obstacle-level velocity (L velocity) output by the lidar sensor, and the point cloud alignment at obstacle-level velocity (R velocity) output by the millimeter-wave radar sensor.

[0151] For different candidate prediction speeds, the overlap C of the point cloud alignment effect under each candidate prediction speed is calculated using the following formula:

[0152]

[0153] Used to represent the three-dimensional coordinate vector and RGB vector of the first point in the first point cloud data. The three-dimensional coordinate vector and RGB vector of the second point in the second point cloud data are used to represent the second point. `mahanalobis()` is used to represent the Mahalanobis distance, where the first and second points correspond to each other. The logarithmic function is L(x) = lg(x + 0.001) - lg(0.001), n i Used to represent the number of points contained in the first point cloud data and the second point cloud data.

[0154] Specifically, during the calculation process, a single sliding window method can be used for one-time calculations to reduce computational complexity by 50%; the sampling rate of the point cloud data can be reduced to reduce computational complexity by 75%; the floating-point precision of the point cloud data can be reduced to reduce computational complexity by 50%; and ground points in the point cloud data can be filtered to reduce computational complexity by 10%. This reduces the amount of computation, improves computational efficiency, and reduces the consumption of computing resources.

[0155] Based on the overlap C corresponding to each candidate prediction speed, the optimal solution, i.e. the target prediction speed, is determined from multiple candidate prediction speeds.

[0156] The velocity prediction method according to embodiments of this disclosure combines point cloud data from a lidar sensor or radar sensor with RGB data from a camera sensor. By utilizing the raw sensor data, it performs high-dimensional, fine-grained motion estimation, solving the technical problem of low accuracy in motion estimation during low-speed motion scenarios. Furthermore, the raw sensor data allows for simultaneous velocity prediction and confidence level calculation, balancing computational efficiency and accuracy.

[0157] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0158] According to embodiments of this disclosure, this disclosure also provides a speed prediction device.

[0159] like Figure 12 As shown, the speed prediction device includes:

[0160] The acquisition module 1201 is used to acquire signal source data of the target object and multiple candidate prediction velocities. The signal source data includes raw data from at least one sensor.

[0161] The confidence calculation module 1202 is used to calculate the confidence of each candidate prediction velocity using raw data from at least one sensor.

[0162] The target prediction speed determination module 1203 is used to determine the target prediction speed of the target object from multiple candidate prediction speeds based on the confidence level of each candidate prediction speed.

[0163] In one implementation, the sensor's raw data includes point cloud data at at least two different reference times; the confidence calculation module 1202 includes:

[0164] The reference speed to determination submodule is used to determine the reference speed value corresponding to the candidate prediction speed at two different reference times for each candidate prediction speed;

[0165] The translation relationship determination submodule determines the translation relationship between two point cloud data based on two reference velocity values;

[0166] The overlap determination submodule is used to determine the overlap between two point cloud data based on the translation relationship.

[0167] The confidence level determination submodule is used to determine the confidence level of the candidate prediction velocity based on the overlap.

[0168] In one implementation, the translation relationship determination submodule includes:

[0169] The interpolation unit is used to interpolate two reference velocity values ​​to obtain the average velocity value of the two reference velocity values;

[0170] The translation relationship determination unit is used to determine the translation relationship between two point cloud data based on the average velocity value and the time difference between two different reference times.

[0171] In one implementation, the overlap determination submodule includes:

[0172] The alignment unit is used to align two point cloud data according to the translation relationship, and obtain the alignment result of the two point cloud data. The alignment result includes multiple corresponding point groups of the two point cloud data.

[0173] The calculation unit is used to calculate the logarithmic value of the Mahalanobis distance for each of the multiple corresponding point groups;

[0174] The overlap determination unit is used to determine the overlap of two point cloud data based on the logarithmic value of the Mahalanobis distance of each of the multiple corresponding point groups.

[0175] In one embodiment, the acquisition module 1201 includes:

[0176] The first acquisition submodule is used to determine, based on the point cloud data from the lidar sensor, the first preprocessing prediction speed and the first postprocessing prediction speed corresponding to the target object, wherein the postprocessing prediction speed is obtained by postprocessing based on the preprocessing prediction speed; and,

[0177] The second acquisition submodule is used to determine, based on the point cloud data from the millimeter-wave radar sensor, the second preprocessing prediction speed and the second post-processing prediction speed corresponding to the target object, wherein the second post-processing prediction speed is obtained by post-processing based on the second preprocessing prediction speed; and

[0178] The third acquisition submodule is used to determine the third preprocessing prediction speed and the horizontal splitting prediction speed corresponding to the target object based on the RGB data of the camera sensor. The horizontal splitting prediction speed is obtained by horizontal splitting based on the third preprocessing prediction speed.

[0179] The candidate prediction speed determination submodule is used to determine at least two of the following: the first preprocessing prediction speed, the first postprocessing prediction speed, the second preprocessing prediction speed, the second postprocessing prediction speed, the third preprocessing prediction speed, and the lateral splitting prediction speed as at least two candidate prediction speeds.

[0180] In one embodiment, the acquisition module 1201 further includes:

[0181] The fourth acquisition submodule is used to perform splicing processing on at least two of the first post-processing prediction speed, the second post-processing prediction speed, and the horizontal splitting prediction speed to obtain at least one splicing speed.

[0182] The candidate prediction speed determination submodule is also used to determine at least one splicing speed as at least one candidate prediction speed among a plurality of candidate prediction speeds.

[0183] In one embodiment, the acquisition module 1201 further includes:

[0184] The fifth acquisition submodule is used to acquire the historical motion characteristics of the target object; and, based on the historical motion characteristics, to calculate the compensation velocity of the target object;

[0185] The candidate prediction speed determination submodule is also used to determine the compensation speed of the target object as one of a number of candidate prediction speeds.

[0186] In one implementation, the signal source data further includes roadside sensing data; the acquisition module 1201 further includes:

[0187] The sixth acquisition submodule is used to determine the roadside sensing speed of the target object based on the roadside sensing data;

[0188] The candidate predicted speed determination submodule is also used to determine the roadside perceived speed as one of a number of candidate predicted speeds.

[0189] In one embodiment, the acquisition module 1201 includes:

[0190] The first prediction velocity determination submodule is used to determine multiple first prediction velocities corresponding to the target object based on signal source data; and,

[0191] The second prediction speed determination submodule is used to obtain at least one second prediction speed based on at least one first prediction speed using an optimization algorithm;

[0192] The candidate prediction velocity determination submodule is used to determine at least one first prediction velocity and / or at least one second prediction velocity as at least one candidate prediction velocity among a plurality of candidate prediction velocities.

[0193] In one implementation, the second predicted velocity determination submodule is further configured to:

[0194] Based on at least one first predicted velocity, at least one second predicted velocity is obtained using a neighborhood search algorithm and / or a local minimum algorithm.

[0195] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0196] Figure 13A schematic block diagram of an example electronic device 1300 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0197] like Figure 13 As shown, device 1300 includes a computing unit 1301, which can perform various appropriate actions and processes according to a computer program stored in read-only memory (ROM) 1302 or a computer program loaded from storage unit 1308 into random access memory (RAM) 1303. The RAM 1303 may also store various programs and data required for the operation of device 1300. The computing unit 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. Input / output (I / O) interface 1305 is also connected to bus 1304.

[0198] Multiple components in device 1300 are connected to I / O interface 1305, including: input unit 1306, such as keyboard, mouse, etc.; output unit 1307, such as various types of monitors, speakers, etc.; storage unit 1308, such as disk, optical disk, etc.; and communication unit 1309, such as network card, modem, wireless transceiver, etc. Communication unit 1309 allows device 1300 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0199] The computing unit 1301 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 1301 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 1301 performs the various methods and processes described above, such as the speed prediction method. For example, in some embodiments, the speed prediction method may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 1308. In some embodiments, part or all of the computer program may be loaded and / or installed on device 1300 via ROM 1302 and / or communication unit 1309. When the computer program is loaded into RAM 1303 and executed by the computing unit 1301, one or more steps of the speed prediction method described above may be performed. Alternatively, in other embodiments, the computing unit 1301 may be configured to perform the speed prediction method by any other suitable means (e.g., by means of firmware).

[0200] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transferring data and instructions to the storage system, the at least one input device, and the at least one output device.

[0201] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0202] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0203] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0204] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with embodiments of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0205] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0206] According to embodiments of this disclosure, this disclosure also provides an autonomous driving vehicle, including a speed acquisition device and / or electronic device according to the above embodiments of this disclosure.

[0207] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0208] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A speed prediction method, comprising: Acquire signal source data of the target object and multiple candidate prediction velocities, wherein the signal source data includes raw data from at least one sensor; and wherein the raw data from the sensor includes point cloud data at at least two different reference times. Using raw data from the at least one sensor, calculate the confidence level for each of the candidate predicted velocities, including: For each of the candidate prediction speeds: Determine the reference velocity values ​​corresponding to the candidate prediction velocity at the two different reference times; The translation relationship between the two point cloud data is determined based on the two reference velocity values. The degree of overlap between the two point cloud data sets is determined based on the translation relationship; The confidence level of the candidate prediction velocity is determined based on the overlap degree. The target predicted velocity of the target object is determined from the plurality of candidate predicted velocities based on the confidence level of each candidate predicted velocity.

2. The method according to claim 1, wherein, Based on the two reference velocity values, the translation relationship between the two point cloud data is determined, including: The two reference velocity values ​​are interpolated to obtain the average velocity value of the two reference velocity values. The translation relationship between the two point cloud data is determined based on the average velocity value and the time difference between the two different reference times.

3. The method according to claim 1, wherein, Determining the overlap between two point cloud datasets based on the translation relationship includes: The two point cloud data are aligned according to the translation relationship to obtain the alignment result of the two point cloud data, and the alignment result includes multiple corresponding point groups of the two point cloud data. Calculate the logarithmic value of the Mahalanobis distance for each of the plurality of corresponding point groups; The overlap between the two point cloud data is determined based on the logarithmic value of the Mahalanobis distance of each of the plurality of corresponding point groups.

4. The method according to claim 1, wherein, Obtaining multiple candidate prediction velocities for the target object, including: Based on the point cloud data from the lidar sensor, a first preprocessing prediction speed and a first post-processing prediction speed corresponding to the target object are determined, wherein the first post-processing prediction speed is obtained by post-processing based on the first preprocessing prediction speed; and Based on point cloud data from a millimeter-wave radar sensor, a second preprocessing prediction speed and a second post-processing prediction speed corresponding to the target object are determined, wherein the second post-processing prediction speed is obtained by post-processing based on the second preprocessing prediction speed; and... Based on the RGB data from the camera sensor, the third preprocessing prediction speed and the lateral splitting prediction speed corresponding to the target object are determined. The lateral splitting prediction speed is obtained by performing lateral splitting based on the third preprocessing prediction speed. At least two of the following are selected as the at least two candidate prediction speeds: the first preprocessing prediction speed, the first postprocessing prediction speed, the second preprocessing prediction speed, the second postprocessing prediction speed, the third preprocessing prediction speed, and the lateral splitting prediction speed.

5. The method according to claim 4, wherein, The method for obtaining multiple candidate prediction speeds for the target object further includes: Based on the first post-processing prediction speed, the second post-processing prediction speed, and the horizontal splitting prediction speed, at least two of them are spliced ​​together to obtain at least one splicing speed. The at least one splicing speed is determined as at least one of the plurality of candidate prediction speeds.

6. The method according to claim 4, wherein, The method for obtaining multiple candidate prediction speeds for the target object further includes: Obtain the historical motion characteristics of the target object; Based on the historical motion characteristics, the compensation velocity of the target object is calculated; The compensation speed of the target object is determined as one of the multiple candidate prediction speeds.

7. The method according to claim 4, wherein, The signal source data also includes roadside sensing data; acquiring multiple candidate predicted speeds for the target object further includes: Based on the roadside sensing data, determine the roadside sensing speed of the target object; The roadside sensing speed is determined as one of a plurality of candidate predicted speeds.

8. The method according to claim 1, wherein, Obtaining multiple candidate prediction velocities for the target object, including: Based on the signal source data, determine multiple first predicted velocities corresponding to the target object; and, Based on at least one first predicted velocity, at least one second predicted velocity is obtained using an optimization algorithm; At least one of the first predicted velocities and / or at least one of the second predicted velocities are determined as at least one of the plurality of candidate predicted velocities.

9. The method according to claim 8, wherein, Based on at least one first predicted velocity, at least one second predicted velocity is obtained using an optimization algorithm, including: Based on at least one of the first predicted velocities, the at least one second predicted velocity is obtained using a neighborhood search algorithm and / or a local minimum algorithm.

10. A speed prediction device, comprising: An acquisition module is used to acquire signal source data of a target object and multiple candidate prediction velocities, wherein the signal source data includes raw data from at least one sensor; and wherein the raw data from the sensor includes point cloud data at at least two different reference times. A confidence calculation module is used to calculate the confidence level of each of the candidate prediction velocities using the raw data from the at least one sensor. The confidence calculation module includes: The reference speed to determination submodule is used to determine the reference speed value corresponding to each candidate prediction speed at the two different reference times for each candidate prediction speed. The translation relationship determination submodule determines the translation relationship between the two point cloud data based on the two reference velocity values. The overlap determination submodule is used to determine the overlap degree of two point cloud data based on the translation relationship. The confidence determination submodule is used to determine the confidence level of the candidate prediction velocity based on the overlap. The target prediction speed determination module is used to determine the target prediction speed of the target object from a plurality of candidate prediction speeds based on the confidence level of each candidate prediction speed.

11. The apparatus according to claim 10, wherein, The translation relationship determination submodule includes: An interpolation unit is used to interpolate two reference velocity values ​​to obtain the average velocity value of the two reference velocity values. The translation relationship determination unit is used to determine the translation relationship between the two point cloud data based on the average velocity value and the time difference between the two different reference times.

12. The apparatus according to claim 10, wherein, The overlap determination submodule includes: An alignment unit is used to align two point cloud data according to the translation relationship to obtain an alignment result of the two point cloud data, the alignment result including multiple corresponding point groups of the two point cloud data; A calculation unit is used to calculate the logarithmic value of the Mahalanobis distance for each of the plurality of corresponding point groups; The overlap determination unit is used to determine the overlap of two point cloud data based on the logarithmic value of the Mahalanobis distance of each of the plurality of corresponding point groups.

13. The apparatus according to claim 10, wherein, The acquisition module includes: The first acquisition submodule is configured to determine, based on point cloud data from a lidar sensor, a first preprocessing prediction speed and a first post-processing prediction speed corresponding to the target object, wherein the first post-processing prediction speed is obtained by post-processing based on the first preprocessing prediction speed; and, The second acquisition submodule is used to determine, based on the point cloud data from the millimeter-wave radar sensor, the second preprocessing prediction speed and the second post-processing prediction speed corresponding to the target object, wherein the second post-processing prediction speed is obtained by post-processing based on the second preprocessing prediction speed; and, The third acquisition submodule is used to determine the third preprocessing prediction speed and the horizontal splitting prediction speed corresponding to the target object based on the RGB data of the camera sensor. The horizontal splitting prediction speed is obtained by horizontal splitting based on the third preprocessing prediction speed. The candidate prediction speed determination submodule is used to determine at least two of the following: the first preprocessing prediction speed, the first postprocessing prediction speed, the second preprocessing prediction speed, the second postprocessing prediction speed, the third preprocessing prediction speed, and the lateral splitting prediction speed as at least two candidate prediction speeds.

14. The apparatus according to claim 13, wherein, The acquisition module further includes: The fourth acquisition submodule is used to perform splicing processing on at least two of the first post-processing prediction speed, the second post-processing prediction speed and the horizontal splitting prediction speed to obtain at least one splicing speed. The candidate prediction speed determination submodule is further configured to determine the at least one splicing speed as at least one of the multiple candidate prediction speeds.

15. The apparatus according to claim 13, wherein, The acquisition module further includes: The fifth acquisition submodule is used to acquire the historical motion characteristics of the target object; and to calculate the compensation speed of the target object based on the historical motion characteristics; The candidate prediction speed determination submodule is further configured to determine the compensation speed of the target object as one of the candidate prediction speeds among a plurality of candidate prediction speeds.

16. The apparatus according to claim 13, wherein, The signal source data also includes roadside sensing data; the acquisition module further includes: The sixth acquisition submodule is used to determine the roadside sensing speed of the target object based on the roadside sensing data; The candidate predicted speed determination submodule is also used to determine the roadside perceived speed as one of the multiple candidate predicted speeds.

17. The apparatus according to claim 10, wherein, The acquisition module includes: The first predicted velocity determination submodule is used to determine multiple first predicted velocities corresponding to the target object based on the signal source data; and, The second prediction speed determination submodule is used to obtain at least one second prediction speed based on at least one first prediction speed using an optimization algorithm; A candidate prediction speed determination submodule is used to determine at least one of the first prediction speeds and / or at least one of the second prediction speeds as at least one of the plurality of candidate prediction speeds.

18. The apparatus according to claim 17, wherein, The second predicted velocity determination submodule is also used for: Based on at least one of the first predicted velocities, the at least one second predicted velocity is obtained using a neighborhood search algorithm and / or a local minimum algorithm.

19. An electronic device comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1 to 9.

20. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1 to 9.

21. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1 to 9.

22. An autonomous vehicle comprising the device of any one of claims 10 to 18 and / or the electronic device of claim 19.

Images