Method and device for calculating acceleration of point cloud labeling box in automatic driving

Abstract

The invention relates to the technical field of data processing, and discloses a method for calculating the acceleration of a point cloud labeling box in automatic driving, and the method comprises the steps: S10, extracting point cloud boxes with the same serial number in a current frame, a previous frame and a next frame from labeling truth value data, and respectively recording the coordinates of the point cloud boxes as Pc1, Pc2 and Pc3; s20, converting a previous frame coordinate Pc1 into a world coordinate system through a car2gnss matrix and a gnss2word matrix of the first frame, and then converting the previous frame coordinate Pc1 into a current frame own vehicle coordinate system through a world2gnss matrix and a gnss2car matrix of the second frame to obtain a converted coordinate Pc1 '; s30, converting the next frame of coordinate Pc3 into a world coordinate system through a car2gnss matrix and a gnss2word matrix of a third frame, and then converting the next frame of coordinate Pc3 into a current frame of own vehicle coordinate system through a world2gnss matrix and a gnss2car matrix of a second frame to obtain a converted coordinate Pc3 '; and S40, on the basis of the converted coordinates Pc1 ', Pc2 and Pc3', component accelerations of the x axis, the y axis and the z axis of the current frame are calculated according to a differential common 7 formula, and the point cloud labeling box acceleration is obtained on the basis of the component accelerations.

Description

Technical Field

[0001] The present invention relates to the field of data processing technology, and in particular to a method and device for calculating the acceleration of a point cloud annotation box in autonomous driving. Background Art

[0002] Point cloud fusion annotation technology plays a crucial role in optimizing autonomous driving algorithms. The ground truth of the point cloud bounding boxes in the point cloud annotation results contains rich spatial information. Acceleration, a key parameter describing the changing state of an object, is crucial for understanding the dynamic behavior of autonomous driving objects, optimizing algorithm models, and improving system performance.

[0003] In point cloud fusion annotation, it is necessary to calculate the acceleration of the point cloud box of the ground truth annotation result. The traditional method of calculating acceleration is to calculate acceleration based on the collected velocity information, which is dependent on the collected data information. Another method is to convert all the annotated data to the world coordinate system and calculate the acceleration information of the box in the world coordinate system. The first method relies on the accuracy of the collected information. If the collected information is inaccurate or the collected data is lost at a certain point in time, the calculated acceleration will be inaccurate. The second method calculates the point cloud acceleration in the world coordinate system. The acceleration value remains unchanged, but the direction of the acceleration is different from the actual direction.

[0004] Therefore, a more accurate method for calculating the acceleration of point cloud annotation boxes is urgently needed. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a more accurate and comprehensive method for calculating the acceleration of the point cloud annotation box in autonomous driving.

[0006] In order to solve the above technical problems, the present invention provides a method for calculating the acceleration of a point cloud annotation box in autonomous driving, the method comprising:

[0007] S10, extract the point cloud frames with the same sequence number in the current frame, the previous frame, and the next frame from the annotated true value data, and record their coordinates as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3), respectively;

[0008] S20, convert the previous frame coordinate Pc1 to the world coordinate system using the car2gnss matrix and gnss2world matrix of the first frame, and then convert it to the vehicle coordinate system of the current frame using the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc1'(x1', y1', z1');

[0009] S30, convert the next frame coordinate Pc3 to the world coordinate system using the car2gnss matrix and gnss2world matrix of the third frame, and then convert it to the vehicle coordinate system of the current frame using the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc3'(x3', y3', z3');

[0010] S40, based on the converted coordinates Pc1'(x1',y1',z1'), Pc2(x2,y2,z2), Pc3'(x3',y3',z3'), calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame respectively according to the differential formula, and obtain the point cloud annotation box acceleration based on the component accelerations.

[0011] Furthermore, the difference calculation formula is:

[0012] ax=(x3'+x1'-2x2) / (t 2 );

[0013] ay=(y3'+y1'-2y2) / (t 2 );

[0014] az=(z3'+z1'-2z2) / (t 2 );

[0015] Where t is the interval time, ax, ay, and az are the component accelerations of the x-axis, y-axis, and z-axis of the current frame, respectively.

[0016] Furthermore, the method comprises

[0017] S50, performing interpolation calculation according to the number of frames in all frames;

[0018] If the number of frames is ≤3, no interpolation calculation is performed on the acceleration of the first and last frames, and the acceleration value of the middle frame is directly assigned to the frames in the first and last frames;

[0019] If the number of frames is greater than 3, the frames of the first and last frames need to be interpolated.

[0020] Furthermore, when the first frame is interpolated, the following steps are performed:

[0021] Get the acceleration values ​​of the box in the second and third frames, recorded as a2(x2,y2,z2) and a3(x3,y3,z3) respectively;

[0022] The acceleration values ​​of the frames in the second and third frames are calculated to obtain the acceleration a1 of the first frame. The specific formula for calculating the acceleration values ​​of the frames in the second and third frames is:

[0023]

[0024] Furthermore, when the last frame is interpolated, the following steps are performed:

[0025] Get the acceleration of the frame of the second to last frame and the third to last frame, recorded as a2(x2,y2,z2) and a3(x3,y3,z3) respectively,

[0026] The acceleration values ​​of the frames in the second and third frames are calculated to obtain the acceleration a1 of the first frame. The specific formula for calculating the acceleration values ​​of the frames in the second and third frames is:

[0027]

[0028] Furthermore, the interval time t is determined by dynamic timestamp alignment, specifically including:

[0029] If the time intervals between adjacent frames are non-uniform, the time axis is resampled using cubic spline interpolation;

[0030] If there is frame loss, the coordinate data of the missing frame is compensated by forward filling and linear interpolation.

[0031] According to another aspect of the present invention, a device for calculating the acceleration of a point cloud annotation box in autonomous driving is provided, the device comprising:

[0032] A data extraction module is used to extract the point cloud frames with the same sequence number in the current frame, the previous frame, and the next frame from the annotated true value data, and record their coordinates as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3) respectively;

[0033] A first coordinate system conversion module is used to convert the coordinates Pc1 of the previous frame to the world coordinate system through the car2gnss matrix and gnss2world matrix of the first frame, and then convert it to the vehicle coordinate system of the current frame through the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinates Pc1'(x1', y1', z1');

[0034] The second coordinate system conversion module is used to convert the next frame coordinate Pc3 to the world coordinate system through the car2gnss matrix and gnss2world matrix of the third frame, and then convert it to the current frame self-vehicle coordinate system through the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc3'(x3',y3',z3');

[0035] An acceleration calculation module is used to calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame based on the converted coordinates Pc1'(x1', y1', z1'), Pc2(x2, y2, z2), and Pc3'(x3', y3', z3') according to the differential formula, and obtain the acceleration of the point cloud annotation box based on the component accelerations.

[0036] Furthermore, the apparatus includes an interpolation processing module, the interpolation processing module being configured to perform interpolation calculations based on the number of frames in all frames;

[0037] If the number of frames is ≤3, no interpolation calculation is performed on the acceleration of the first and last frames, and the acceleration value of the middle frame is directly assigned to the frames in the first and last frames;

[0038] If the number of frames is greater than 3, the frames of the first and last frames need to be interpolated.

[0039] Compared with the prior art, the method for calculating the acceleration of a point cloud annotation box in autonomous driving provided by the present invention has the following advantages:

[0040] This method extracts point cloud frame coordinates from the annotated ground-truth data and, through a series of coordinate transformations, unifies the point cloud frame coordinates of different frames into the current frame's ego-vehicle coordinate system. Acceleration is then calculated using a differential formula based on these transformed coordinates. This method avoids direct reliance on collected velocity data, reduces the impact of acquisition errors, and thus provides a more accurate acceleration calculation method. Furthermore, interpolation is performed on the first and last frames, where acceleration cannot be calculated, enriching the attribute information of the autonomous driving annotated frame's ground-truth value and enhancing the integrity of the annotated data. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Figure 1 This is a flow chart of a method for calculating the acceleration of a point cloud annotation box in autonomous driving provided by an embodiment of the present invention;

[0042] Figure 2 This is a flowchart of a method for calculating the acceleration of a point cloud annotation box in autonomous driving provided by another embodiment of the present invention;

[0043] Figure 3 This is a flowchart of a specific implementation of step S50 provided in an embodiment of the present invention;

[0044] Figure 4 is a flowchart of another specific implementation of step S50 provided in an embodiment of the present invention;

[0045] Figure 5 2 is a schematic diagram of a device for calculating the acceleration of a point cloud annotation box in autonomous driving provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The following description of exemplary embodiments of the present invention is provided in conjunction with the accompanying drawings, in which various details of the embodiments of the present invention are included to facilitate understanding. These details should be considered as merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications may be made to the embodiments described herein without departing from the scope of the present invention. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0047] like Figure 1-4 As shown, in an optional embodiment of the present invention, the method for calculating the acceleration of the point cloud annotation box in the autonomous driving includes:

[0048] S10, extract the point cloud frames with the same sequence number in the current frame, the previous frame, and the next frame from the annotated true value data, and record their coordinates as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3), respectively;

[0049] S20, convert the previous frame coordinate Pc1 to the world coordinate system using the car2gnss matrix and gnss2world matrix of the first frame, and then convert it to the vehicle coordinate system of the current frame using the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc1'(x1', y1', z1');

[0050] S30, convert the next frame coordinate Pc3 to the world coordinate system using the car2gnss matrix and gnss2world matrix of the third frame, and then convert it to the vehicle coordinate system of the current frame using the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc3'(x3', y3', z3');

[0051] S40, based on the converted coordinates Pc1'(x1',y1',z1'), Pc2(x2,y2,z2), Pc3'(x3',y3',z3'), calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame respectively according to the differential formula, and obtain the point cloud annotation box acceleration based on the component accelerations.

[0052] Among them, labeled true value data: In the field of autonomous driving, accurate data usually obtained through manual labeling or specific algorithm labeling is used to verify and train autonomous driving related algorithms. Here it refers to a precise data set containing information such as point cloud frame coordinates. car2gnss matrix: The conversion matrix from the vehicle coordinate system to the global navigation satellite system (GNSS) coordinate system, which is used to convert coordinates in the vehicle coordinate system to coordinates in the GNSS coordinate system. world2gnss matrix: The conversion matrix from the world coordinate system to the GNSS coordinate system, which is used to convert coordinates in the world coordinate system to coordinates in the GNSS coordinate system.

[0053] Specifically, S10 extracts the point cloud frames with the same serial number in the current frame, previous frame and next frame from the annotated true value data, and records their coordinates as Pc1 (x1, y1, z1), Pc2 (x2, y2, z2), and Pc3 (x3, y3, z3), respectively; the purpose of this step is to obtain the basic data for calculating acceleration, that is, the coordinate information of the point cloud frames with the same serial number in different frames. Providing accurate initial position data for subsequent coordinate conversion and acceleration calculation to ensure the accuracy of the calculation depends on accurate annotated true values. Implementation method: Read the annotated true value data file, which usually stores the coordinate information of the point cloud frame in a specific format (such as JSON, XML, etc.). According to the frame number information, locate the data area of ​​the current frame, previous frame and next frame respectively. In each frame of data, find the point cloud frame with the same serial number and extract its x, y, and z coordinate values, which are recorded as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3), respectively.

[0054] Specifically, in step S20, the previous frame coordinates Pc1 are converted to the world coordinate system using the car2gnss matrix and gnss2world matrix of the first frame, and then converted to the current frame's ego-vehicle coordinate system using the world2gnss matrix and gnss2car matrix of the second frame, obtaining the converted coordinates Pc1'(x1', y1', z1'). This step converts the point cloud frame coordinates of the previous frame to the current frame's ego-vehicle coordinate system so that calculations can be performed in the same coordinate system as the current frame's coordinates. This eliminates the differences in coordinate systems between different frames, allowing subsequent acceleration calculations to be performed based on a unified coordinate system, improving the accuracy and consistency of the calculations.

[0055] Specifically, in step S30, the next frame coordinate Pc3 is converted to the world coordinate system using the car2gnss matrix and gnss2world matrix of the third frame, and then converted to the current frame's ego-vehicle coordinate system using the world2gnss matrix and gnss2car matrix of the second frame, obtaining the converted coordinate Pc3'(x3', y3', z3'). This step converts the point cloud frame coordinates of the next frame to the current frame's ego-vehicle coordinate system, unifying the coordinate system with the converted coordinates of the current and previous frames. Similar to step S20, this ensures that all coordinates used for calculation are in the same coordinate system, providing a consistent data foundation for subsequent acceleration calculations.

[0056] Specifically, S40, based on the converted coordinates Pc1'(x1', y1', z1'), Pc2(x2, y2, z2), Pc3'(x3', y3', z3'), calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame respectively according to the differential formula, and obtain the acceleration of the point cloud annotation box based on the component accelerations. The purpose of this step is to calculate the component accelerations of the point cloud annotation box on the x, y, and z axes of the current frame according to the converted coordinates, and to obtain the final acceleration of the point cloud annotation box. The acceleration components of the point cloud annotation box on the three axes are accurately calculated through the differential formula, and then comprehensive acceleration information is obtained to provide accurate motion state data for autonomous driving.

[0057] Specifically, combined Figure 1-4 Let's further explain with a complete example:

[0058] S10. Based on the actual labeled true value data, the boxes with the same serial number are obtained from the current frame, the previous frame, and the next frame. We record the coordinates of the box in the current frame as Pc2 (x2, y2, z2), the coordinates of the box in the previous frame as Pc1 (x1, y1, z1), and the coordinates of the box in the next frame as Pc3 (x3, y3, z3).

[0059] S20. Convert Pc1(x1, y1, z1) from the ego-vehicle coordinate system of the previous frame to the world coordinate system using the car2gnss matrix of the first frame and the gnss2world matrix of the first frame. Then, use the world2gnss and gnss2car of the second frame to convert it to the ego-vehicle coordinate system of the second frame. The converted coordinates are Pc2(x1, y1, z1).

[0060] S30. Similarly, the coordinates of the third frame are transferred to the coordinates of the second frame to obtain the converted coordinates Pc2 (x3, y3, z3).

[0061] S40. Using the coordinates of the three center points, Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3), calculate the x-axis acceleration as ax = (x3 + x1 - 2*x2) / (t*t), where x3 is the x-coordinate of the third frame, x2 is the x-coordinate of the second frame, x1 is the x-coordinate of the first frame, and t is the time difference between the two frames. Similarly, the y-axis acceleration ay = (y3 + y1 - 2*y2) / (t*t) and the z-axis acceleration az = (z3 + z1 - 2*y2) / (t*t) can be calculated.

[0062] Combine Figure 2-4 This embodiment also includes:

[0063] S50. Calculate the number of frames in all frames. If the number of frames is less than 3, do not interpolate the acceleration of the first and last frames, and directly assign the acceleration value of the middle frame to the frames in the first and last frames. If the number of frames is greater than 3, interpolate the frames in the first and last frames.

[0064] S60. Perform interpolation on the first frame. Obtain the acceleration values ​​of the frames in the second and third frames, denoted as a2(x2, y2, z2) and a3(x3, y3, z3) respectively. First, subtract the two accelerations by a3-a2, and then take the square root of the obtained value. Then subtract the square root value from the acceleration of the second frame to get the acceleration of the first frame. For example, if the acceleration of the second frame is a2(8,0,0), and the acceleration of the third frame is a3(12,0,0), then the acceleration of the first frame is Then the acceleration of the first frame is a1(6,0,0).

[0065] S70. Perform interpolation on the last frame. Obtain the accelerations of the second-to-last frame and the third-to-last frame, denoted as a2(x2, y2, z2) and a3(x3, y3, z3), respectively. Subtract the two accelerations, but in the opposite order for the last frame: a2 - a3. Then take the square root of the obtained value. Then add the square root value to the acceleration of the penultimate frame to get the acceleration of the first frame. For example, the acceleration of the second-to-last frame is a2(12,0,0), and the acceleration of the third-to-last frame is a3(8,0,0), then the acceleration of the first frame is Then the acceleration of the first frame is a1(14,0,0).

[0066] In an embodiment of the present invention, the point cloud frame coordinates are extracted from the annotated ground truth data. Through a series of coordinate transformation operations, the point cloud frame coordinates of different frames are unified into the ego-vehicle coordinate system of the current frame. Acceleration is then calculated using a differential formula based on the transformed coordinates. This method avoids direct reliance on acquired velocity data, reduces the impact of acquisition errors, and thus provides a more accurate acceleration calculation method. Furthermore, interpolation is performed on the first and last frames, where acceleration cannot be calculated, enriching the attribute information of the autonomous driving annotated frame ground truth, thus enhancing the integrity of the annotated data.

[0067] In an optional embodiment of the present invention, the difference calculation formula is:

[0068] ax=(x3'+x1'-2x2) / (t 2 );

[0069] ay=(y3'+y1'-2y2) / (t 2 );

[0070] az=(z3'+z1'-2z2) / (t 2 );

[0071] Where t is the interval time, ax, ay, and az are the component accelerations of the x-axis, y-axis, and z-axis of the current frame, respectively.

[0072] In an embodiment of the present invention, by using the position information of multiple frames for calculation, the noise interference that may exist in the single-frame data can be reduced to a certain extent. Because noise is often random, the comprehensive calculation of multiple frames of data can smooth the influence of the noise and make the calculated acceleration more accurate. The coordinates used in the formula are precise coordinates after coordinate conversion. These coordinates are derived from the labeled true value data and have high accuracy. Calculating the acceleration based on these precise coordinates can ensure the reliability of the calculation results. Accurate x-, y-, and z-axis component acceleration information can help the autonomous driving system more accurately judge the motion state of surrounding objects (such as other vehicles, pedestrians, etc.). For example, by analyzing the magnitude and direction of the acceleration, it can be determined whether the object is accelerating, decelerating, or moving at a constant speed, as well as the change in the direction of motion.

[0073] like Figure 3-4 As shown, in an optional embodiment of the present invention, the method includes

[0074] S50, performing interpolation calculation according to the number of frames in all frames;

[0075] If the number of frames is ≤3, no interpolation calculation is performed on the acceleration of the first and last frames, and the acceleration value of the middle frame is directly assigned to the frames in the first and last frames;

[0076] If the number of frames is greater than 3, the frames of the first and last frames need to be interpolated.

[0077] In an optional embodiment of the present invention, when the first frame is interpolated, the following steps are performed:

[0078] Get the acceleration values ​​of the box in the second and third frames, recorded as a2(x2,y2,z2) and a3(x3,y3,z3) respectively;

[0079] The acceleration values ​​of the frames in the second and third frames are calculated to obtain the acceleration a1 of the first frame. The specific formula for calculating the acceleration values ​​of the frames in the second and third frames is:

[0080]

[0081] like Figure 3-4 As shown, in an optional embodiment of the present invention, when the tail frame is interpolated, the following steps are performed:

[0082] Get the acceleration of the frame of the second to last frame and the third to last frame, recorded as a2(x2,y2,z2) and a3(x3,y3,z3) respectively,

[0083] The acceleration values ​​of the frames in the second and third frames are calculated to obtain the acceleration a1 of the first frame. The specific formula for calculating the acceleration values ​​of the frames in the second and third frames is:

[0084]

[0085] like Figure 5 According to another aspect of the present invention, a device for calculating the acceleration of a point cloud annotation box in autonomous driving is provided, the device comprising:

[0086] A data extraction module is used to extract the point cloud frames with the same sequence number in the current frame, the previous frame, and the next frame from the annotated true value data, and record their coordinates as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3) respectively;

[0087] A first coordinate system conversion module is used to convert the coordinates Pc1 of the previous frame to the world coordinate system through the car2gnss matrix and gnss2world matrix of the first frame, and then convert it to the vehicle coordinate system of the current frame through the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinates Pc1'(x1', y1', z1');

[0088] The second coordinate system conversion module is used to convert the next frame coordinate Pc3 to the world coordinate system through the car2gnss matrix and gnss2world matrix of the third frame, and then convert it to the current frame self-vehicle coordinate system through the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc3'(x3',y3',z3');

[0089] An acceleration calculation module is used to calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame based on the converted coordinates Pc1'(x1', y1', z1'), Pc2(x2, y2, z2), and Pc3'(x3', y3', z3') according to the differential formula, and obtain the acceleration of the point cloud annotation box based on the component accelerations.

[0090] This method extracts point cloud frame coordinates from the annotated ground-truth data and, through a series of coordinate transformations, unifies the point cloud frame coordinates of different frames into the current frame's ego-vehicle coordinate system. Acceleration is then calculated using a differential formula based on these transformed coordinates. This method avoids direct reliance on collected velocity data, reduces the impact of acquisition errors, and thus provides a more accurate acceleration calculation method. Furthermore, interpolation is performed on the first and last frames, where acceleration cannot be calculated, enriching the attribute information of the autonomous driving annotated frame's ground-truth value and enhancing the integrity of the annotated data.

[0091] In an optional embodiment of the present invention, the apparatus includes an interpolation processing module, the interpolation processing module being configured to perform interpolation calculations based on the number of frames in all frames;

[0092] If the number of frames is ≤3, no interpolation calculation is performed on the acceleration of the first and last frames, and the acceleration value of the middle frame is directly assigned to the frames in the first and last frames;

[0093] If the number of frames is greater than 3, the frames of the first and last frames need to be interpolated.

[0094] In this embodiment of the present invention, through interpolation processing, acceleration data can better reflect the actual motion state of objects, especially when there are a large number of frames, reducing sudden changes caused by missing data or noise. Smooth and continuous acceleration data helps the autonomous driving system more accurately judge the motion trends of surrounding objects, thereby making more reliable decisions and improving system safety and stability. Selecting different processing methods based on the number of frames ensures computational efficiency in simple scenarios while meeting the high-precision requirements in complex ones.

[0095] It should be understood that the various forms of the processes shown above can be used to reorder, add, or delete steps. For example, the steps described in the present invention can be performed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present invention can be achieved. This is not limited herein.

[0096] The above specific embodiments do not limit the scope of protection of the present invention. Those skilled in the art will appreciate that various modifications, combinations, sub-combinations, and substitutions may be made based on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the principles of the present invention are intended to be included within the scope of protection of the present invention.

Claims

1. A method for calculating the acceleration of a point cloud annotation box in autonomous driving, characterized in that: The method comprises: S10, extract the point cloud frames with the same sequence number in the current frame, the previous frame, and the next frame from the annotated true value data, and record their coordinates as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3), respectively; S20, convert the previous frame coordinate Pc1 to the world coordinate system using the car2gnss matrix and gnss2world matrix of the first frame, and then convert it to the vehicle coordinate system of the current frame using the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc1'(x1', y1', z1'); S30, convert the next frame coordinate Pc3 to the world coordinate system using the car2gnss matrix and gnss2world matrix of the third frame, and then convert it to the vehicle coordinate system of the current frame using the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc3'(x3', y3', z3'); S40, based on the converted coordinates Pc1'(x1',y1',z1'), Pc2(x2,y2,z2), Pc3'(x3',y3',z3'), calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame respectively according to the differential formula, and obtain the point cloud annotation box acceleration based on the component accelerations.

2. The method for calculating the acceleration of a point cloud annotation box in autonomous driving according to claim 1, characterized in that: The difference calculation formula is: ax=(x3'+x1'-2x2) / (t 2 ); <h2 style=";text-align:left;direction:ltr">ay=(y3'+y1'-2y2) / (t<h2 style=";text-align:left;direction:ltr"> 2 <h2 style=";text-align:left;direction:ltr"> ); az=(z3'+z1'-2z2) / (t 2 ); Where t is the interval time, ax, ay, and az are the component accelerations of the x-axis, y-axis, and z-axis of the current frame, respectively.

3. The method for calculating the acceleration of a point cloud annotation box in autonomous driving according to claim 2, characterized in that: The method comprises S50, performing interpolation calculation according to the number of frames in all frames; If the number of frames is ≤3, no interpolation calculation is performed on the acceleration of the first and last frames, and the acceleration value of the middle frame is directly assigned to the frames in the first and last frames; If the number of frames is greater than 3, the frames of the first and last frames need to be interpolated.

4. The method for calculating the acceleration of a point cloud annotation box in autonomous driving according to claim 3, characterized in that: When interpolation is performed on the first frame, the following steps are performed: Get the acceleration values ​​of the box in the second and third frames, recorded as a2(x2,y2,z2) and a3(x3,y3,z3) respectively; The acceleration values ​​of the frames in the second and third frames are calculated to obtain the acceleration a1 of the first frame; wherein the specific formula for calculating the acceleration values ​​of the frames in the second and third frames is:

5. The method for calculating the acceleration of a point cloud annotation box in autonomous driving according to claim 4, characterized in that: When the last frame is interpolated, the following steps are performed: Get the acceleration of the frame of the second to last frame and the third to last frame, recorded as a2(x2,y2,z2) and a3(x3,y3,z3) respectively, The acceleration values ​​of the frames in the second and third frames are calculated to obtain the acceleration a1 of the first frame. The specific formula for calculating the acceleration values ​​of the frames in the second and third frames is:

6. The method according to claim 2, characterized in that The interval time t is determined by dynamic timestamp alignment, specifically including: If the time intervals between adjacent frames are non-uniform, the time axis is resampled using cubic spline interpolation; If there is frame loss, the coordinate data of the missing frame is compensated by forward filling and linear interpolation.

7. A device for calculating the acceleration of a point cloud annotation box in autonomous driving, characterized in that: The device comprises: A data extraction module is used to extract the point cloud frames with the same sequence number in the current frame, the previous frame, and the next frame from the annotated true value data, and record their coordinates as Pc1(x1, y1, z1), Pc2(x2, y2, z2), and Pc3(x3, y3, z3) respectively; A first coordinate system conversion module is used to convert the coordinates Pc1 of the previous frame to the world coordinate system through the car2gnss matrix and gnss2world matrix of the first frame, and then convert it to the vehicle coordinate system of the current frame through the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinates Pc1'(x1', y1', z1'); The second coordinate system conversion module is used to convert the next frame coordinate Pc3 to the world coordinate system through the car2gnss matrix and gnss2world matrix of the third frame, and then convert it to the current frame self-vehicle coordinate system through the world2gnss matrix and gnss2car matrix of the second frame to obtain the converted coordinate Pc3'(x3',y3',z3'); An acceleration calculation module is used to calculate the component accelerations of the x-axis, y-axis, and z-axis of the current frame based on the converted coordinates Pc1'(x1', y1', z1'), Pc2(x2, y2, z2), and Pc3'(x3', y3', z3') according to the differential formula, and obtain the acceleration of the point cloud annotation box based on the component accelerations.

8. The device according to claim 7, characterized in that The device includes an interpolation processing module, which is used to perform interpolation calculations based on the number of frames in all frames; If the number of frames is ≤3, no interpolation calculation is performed on the acceleration of the first and last frames, and the acceleration value of the middle frame is directly assigned to the frames in the first and last frames; If the number of frames is greater than 3, the frames of the first and last frames need to be interpolated.

Images