A point cloud data processing method and a training data processing method and device

Abstract

Embodiments of the present application provide a point cloud data processing method, a training data processing method and device. In the point cloud data processing method, the point cloud data and initialization bounding box information of the point cloud data are obtained; the initialization bounding box information includes N initialization bounding boxes and position information of each initialization bounding box, the N initialization bounding boxes are used to divide the point cloud data into N initialization point cloud sets, and N is a positive integer; a target conversion matrix of the point cloud data is determined, the initialization bounding box information is corrected according to the target conversion matrix, target bounding box information of the point cloud data is obtained, the target bounding box information includes N target bounding boxes and position information of each target bounding box, and the N target bounding boxes divide the point cloud data into N target point cloud sets. Through the method, the accuracy of the determined target bounding box information is ensured while improving the point cloud data processing efficiency.

Claims

1. A point cloud data processing method, characterized in that, The method includes: Acquire point cloud data and initialization detection box information of the point cloud data; wherein, the initialization detection box information includes N initialization detection boxes and position information of each initialization detection box, the N initialization detection boxes are used to divide the point cloud data into N initialization point cloud sets, where N is a positive integer; A target transformation matrix is ​​determined for the point cloud data. The target transformation matrix is ​​used to correct the initial detection box information. The target transformation matrix is ​​determined based on the reference detection box information and the initial detection box information of the point cloud data. The reference detection box information is obtained based on the point cloud data and is used to provide reference information for correcting the initial detection box. The initial detection box information is corrected according to the target transformation matrix to obtain the target detection box information of the point cloud data. The target detection box information includes N target detection boxes and the position information of each target detection box. The N target detection boxes are used to divide the point cloud data into N target point cloud sets.

2. The method as described in claim 1, characterized in that, Determining the target transformation matrix of the point cloud data includes: The reference detection box information of the point cloud data is determined based on a set perception model. The reference detection box information includes S reference detection boxes and the position information and confidence level of each reference detection box. The S reference detection boxes are used to divide the point cloud data into S reference point cloud sets, where S is a positive integer. The set perception model is used to characterize the correspondence between the point cloud data and the detection box information. M reference detection boxes are determined from the S reference detection boxes, and the confidence level of any one of the M reference detection boxes is less than a first preset threshold; where M is a positive integer less than or equal to S. Select M initialization detection boxes from the N initialization detection boxes that match the M reference detection boxes; A first transformation matrix is ​​determined based on the position information of the M initial detection boxes and the position information of the M reference detection boxes. The first transformation matrix is ​​used to convert the position information of the M initial detection boxes into the position information of the M reference detection boxes. Calculate the difference between each of the at least one specified transformation matrix and the first transformation matrix; The transformation matrix with the smallest difference from the first transformation matrix is ​​taken as the target transformation matrix.

3. The method as described in claim 2, characterized in that, Selecting M initialization detection boxes from the N initialization detection boxes that match the M reference detection boxes includes: Determine the matching parameters of the N initial detection boxes, wherein the matching parameters of any initial detection box include the translation-invariant characteristic parameters and / or rotation-invariant characteristic parameters of the initial detection box; Determine the matching parameters of the M reference detection boxes, wherein the matching parameters of any reference detection box include the translation-invariant characteristic parameters and / or rotation-invariant characteristic parameters of the reference detection box; Among the N initialization detection frames, select the M initialization detection frames whose difference between the matching parameter and the matching parameter of any one of the M reference detection frames is less than a second preset threshold.

4. The method as described in claim 2 or 3, characterized in that, After obtaining the target detection bounding box information of the point cloud data, the method further includes: The point cloud data and the target detection box information are used as a set of training data, which is used to train the target detection model.

5. The method as described in claim 4, characterized in that, Before using the point cloud data and the target detection box information as a set of training data, the method further includes: Obtain the feature parameters of the M reference detection boxes. The feature parameters of any reference detection box are used to represent the complexity of the point cloud distribution of the reference point cloud set corresponding to the reference detection box. The feature parameters of the point cloud data are determined based on the feature parameters of the M reference detection boxes. The feature parameters of the point cloud data are used to represent the complexity of the point cloud distribution in the point cloud data. The data type of the point cloud data is determined based on the feature parameters of the point cloud data; The step of using the point cloud data and the target detection box information as a set of training data includes: The point cloud data and the target detection box information are used as a set of training data corresponding to the data type.

6. A training data processing method, characterized in that, The method includes: Get Q sets of candidate point cloud data and target detection box information for each set of candidate point cloud data. The target detection box information for any set of candidate point cloud data includes at least one target detection box and the position information of each target detection box. The at least one target detection box is used to divide the candidate point cloud data into at least one set of target point clouds. Q is a positive integer. Determine the data type of each group of candidate point cloud data in the Q groups of candidate point cloud data; Based on the data type of each group of candidate point cloud data, determine K target data types, where K target data types include the data type of each group of candidate point cloud data, and K is a positive integer; Select the first point cloud data from T groups from the Q groups of candidate point cloud data as the training set; where T is a positive integer less than Q; The proportion of training data for each of the K target data types in the target training data is determined. The target training data is used to train the target detection model. The proportion of training data for each of the K target data types in the target training data is determined by the error value obtained by training the target detection model based on the point cloud data of each data type in the validation set. Based on the data type of each group of first point cloud data in the training set, and the proportion of training data of each of the K target data types in the target training data, P groups of second point cloud data are determined in the training set. The second point cloud data of the P groups and the target detection box information of the second point cloud data of the P groups are used as the target training data; where P is a positive integer less than T.

7. The method as described in claim 6, characterized in that, Determining the proportion of training data for each of the K target data types in the target training data includes: Select the third point cloud data from group Q as the validation set; where R is a positive integer less than Q; and where the third point cloud data from group R does not have any identical point cloud data with the first point cloud data from group T. The target detection model is trained based on the third point cloud data of each data type in the validation set, and the error value corresponding to the third point cloud data of each data type is determined. The error value corresponding to the third point cloud data of any data type is used to represent the degree of fit of the target detection model to the third point cloud data of that data type. The proportion of training data for each data type in the target training data is determined based on the error value corresponding to the third point cloud data for each data type.

8. The method as described in claim 6 or 7, characterized in that, Determining the data type of each group of candidate point cloud data in the Q groups includes: The following steps are performed on the target point cloud data in the Q groups of candidate point cloud data, wherein the target point cloud data traverses each candidate point cloud data in the Q groups of candidate point cloud data: Determine the feature parameters of at least one target detection box in the target point cloud data, wherein the feature parameters of any target detection box are used to represent the complexity of the point cloud distribution of the target point cloud set corresponding to the target detection box; The feature parameters of the target point cloud data are determined based on the feature parameters of the at least one target detection box, and the feature parameters of the target point cloud data are used to represent the complexity of the point cloud distribution in the target point cloud data; The data type of the target point cloud data is determined based on the feature parameters of the target point cloud data.

9. A point cloud data processing device, characterized in that, Includes an acquisition unit and a processing unit; The acquisition unit is used to acquire point cloud data and initialization detection box information of the point cloud data; wherein, the initialization detection box information includes N initialization detection boxes and position information of each initialization detection box, and the N initialization detection boxes are used to divide the point cloud data into N initialization point cloud sets, where N is a positive integer; The processing unit is used to determine the target transformation matrix of the point cloud data. The target transformation matrix is ​​used to correct the initial detection box information. The target transformation matrix is ​​determined based on the reference detection box information and the initial detection box information of the point cloud data. The reference detection box information is obtained based on the point cloud data and is used to provide reference information for correcting the initial detection box. The processing unit corrects the initial detection box information based on the target transformation matrix to obtain the target detection box information of the point cloud data. The target detection box information includes N target detection boxes and the position information of each target detection box. The N target detection boxes are used to divide the point cloud data into N target point cloud sets.

10. The apparatus as claimed in claim 9, characterized in that, The processing unit is specifically used for: Based on a defined perception model, reference detection box information for the point cloud data is determined. The reference detection box information includes S reference detection boxes and the position information and confidence level of each reference detection box. The S reference detection boxes are used to divide the point cloud data into S reference point cloud sets, where S is a positive integer. The established perception model is used to characterize the correspondence between the point cloud data and the detection box information; M reference detection boxes are determined from the S reference detection boxes, and the confidence level of any one of the M reference detection boxes is less than a first preset threshold; where M is a positive integer less than or equal to S. Select M initialization detection boxes from the N initialization detection boxes that match the M reference detection boxes; A first transformation matrix is ​​determined based on the position information of the M initial detection boxes and the position information of the M reference detection boxes. The first transformation matrix is ​​used to convert the position information of the M initial detection boxes into the position information of the M reference detection boxes. Calculate the difference between each of the at least one specified transformation matrix and the first transformation matrix; The transformation matrix with the smallest difference from the first transformation matrix is ​​taken as the target transformation matrix.

11. The apparatus as claimed in claim 10, characterized in that, The processing unit is specifically used for: Determine the matching parameters of the N initial detection boxes, wherein the matching parameters of any initial detection box include the translation-invariant characteristic parameter and / or rotation-invariant characteristic parameter of the initial detection box; Determine the matching parameters of the M reference detection boxes, wherein the matching parameters of any reference detection box include the translation-invariant characteristic parameters and / or rotation-invariant characteristic parameters of the reference detection box; Among the N initialization detection frames, select the M initialization detection frames whose difference between the matching parameter and the matching parameter of any one of the M reference detection frames is less than a second preset threshold.

12. The apparatus as claimed in claim 10 or 11, characterized in that, The processing unit is also used for: After obtaining the target detection bounding box information of the point cloud data, the point cloud data and the target detection bounding box information are used as a set of training data, which is used to train the target detection model.

13. The apparatus as claimed in claim 12, characterized in that, The processing unit is also used for: Before using the point cloud data and the target detection box information as a set of training data, the feature parameters of the M reference detection boxes are obtained. The feature parameters of any reference detection box are used to represent the complexity of the point cloud distribution of the reference point cloud set corresponding to the reference detection box. The feature parameters of the point cloud data are determined based on the feature parameters of the M reference detection boxes. The feature parameters of the point cloud data are used to represent the complexity of the point cloud distribution in the point cloud data. The data type of the point cloud data is determined based on the feature parameters of the point cloud data; The processing unit is specifically used for: The point cloud data and the target detection box information are used as a set of training data corresponding to the data type.

14. A training data processing apparatus, characterized in that, Includes an acquisition unit and a processing unit; The acquisition unit is used to acquire Q sets of candidate point cloud data and target detection box information for each set of candidate point cloud data. The target detection box information for any set of candidate point cloud data includes at least one target detection box and position information for each target detection box. The at least one target detection box is used to divide the candidate point cloud data into at least one set of target point clouds, where Q is a positive integer. The processing unit is used to determine the data type of each group of candidate point cloud data in the Q groups of candidate point cloud data; determine K target data types based on the data type of each group of candidate point cloud data, wherein the K target data types contain the data type of each group of candidate point cloud data, and K is a positive integer; select T groups of first point cloud data as training set from the Q groups of candidate point cloud data; wherein T is a positive integer less than Q; determine the proportion of training data of each data type in the K target data types in the target training data, wherein the target training data is used to train the target detection model, and the proportion of training data of each data type in the K target data types in the target training data is determined by the error value obtained by training the target detection model based on the point cloud data of each data type in the validation set; determine P groups of second point cloud data in the training set based on the data type of each group of first point cloud data in the training set and the proportion of training data of each data type in the K target data types in the target training data; and use the P groups of second point cloud data and the target detection box information of the P groups of second point cloud data as the target training data; wherein P is a positive integer less than T.

15. The apparatus as claimed in claim 14, characterized in that, The processing unit is specifically used for: Select the third point cloud data from group Q as the validation set; where R is a positive integer less than Q; and where the third point cloud data from group R does not have any identical point cloud data with the first point cloud data from group T. The target detection model is trained based on the third point cloud data of each data type in the validation set, and the error value corresponding to the third point cloud data of each data type is determined. The error value corresponding to the third point cloud data of any data type is used to represent the degree of fit of the target detection model to the third point cloud data of that data type. The proportion of training data for each data type in the target training data is determined based on the error value corresponding to the third point cloud data for each data type.

16. The apparatus as claimed in claim 14 or 15, characterized in that, The processing unit is specifically used for: The following steps are performed on the target point cloud data in the Q groups of candidate point cloud data, wherein the target point cloud data traverses each candidate point cloud data in the Q groups of candidate point cloud data: Determine the feature parameters of at least one target detection box in the target point cloud data, wherein the feature parameters of any target detection box are used to represent the complexity of the point cloud distribution of the target point cloud set corresponding to the target detection box; The feature parameters of the target point cloud data are determined based on the feature parameters of the at least one target detection box, and the feature parameters of the target point cloud data are used to represent the complexity of the point cloud distribution in the target point cloud data; The data type of the target point cloud data is determined based on the feature parameters of the target point cloud data.

17. A point cloud data processing device, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is used to execute the calculation program stored in the memory to implement the method as described in any one of claims 1 to 5.

18. A training data processing apparatus, characterized in that, Including memory and processor; The memory is used to store computer programs; The processor is used to execute the calculation program stored in the memory to implement the method as described in any one of claims 6 to 8.

19. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a data processing device, causes the data processing device to perform the method as described in any one of claims 1 to 5.

20. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when run on a data processing device, causes the data processing device to perform the method as described in any one of claims 6 to 8.

Images