CN113191173A - Training data acquisition method and device - Google Patents

Training data acquisition method and device Download PDF

Info

Publication number
CN113191173A
CN113191173A CN202010038695.8A CN202010038695A CN113191173A CN 113191173 A CN113191173 A CN 113191173A CN 202010038695 A CN202010038695 A CN 202010038695A CN 113191173 A CN113191173 A CN 113191173A
Authority
CN
China
Prior art keywords
data set
data
training
training data
sub
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202010038695.8A
Other languages
Chinese (zh)
Inventor
李万华
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Horizon Robotics Technology Research and Development Co Ltd
Original Assignee
Beijing Horizon Robotics Technology Research and Development Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Horizon Robotics Technology Research and Development Co Ltd filed Critical Beijing Horizon Robotics Technology Research and Development Co Ltd
Priority to CN202010038695.8A priority Critical patent/CN113191173A/en
Publication of CN113191173A publication Critical patent/CN113191173A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
    • G06V20/00—Scenes; Scene-specific elements
    • G06V20/50—Context or environment of the image
    • G06V20/56—Context or environment of the image exterior to a vehicle by using sensors mounted on the vehicle
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W50/0097—Predicting future conditions
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • B—PERFORMING OPERATIONS; TRANSPORTING
    • B60—VEHICLES IN GENERAL
    • B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
    • B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
    • B60W2050/0001—Details of the control system
    • B60W2050/0002—Automatic control, details of type of controller or control system architecture

Abstract

A training data acquisition method and device are disclosed, the method comprises: acquiring an original track data set, wherein the original track data set comprises at least one piece of original track data; mapping the original trajectory data to an occupation grid map, and acquiring corresponding first data to construct a first data set, and acquiring the first data set according to the original data set; processing the first data set according to a first preset mode to obtain a second data set; and carrying out weighted summation on the first data set and the second data set to obtain a training data set. The method and the device ensure that the difficulty degree of the training data in the obtained training data set is greatly improved compared with that of the first data, so that the data quality of the training data set is integrally improved, and the training efficiency of the trajectory prediction network adopting the training data set and the generalization capability of the trajectory prediction network are improved.

Description

Training data acquisition method and device
Technical Field
The present application relates to the field of image processing technologies, and in particular, to a training data acquisition method and apparatus.
Background
In the path planning in the automatic driving field, an automatic driving system needs to predict the track of surrounding obstacle vehicles, and the predicted track is used in a path planning link, so that a path planning algorithm can process complex road conditions. In order to better perform the trajectory prediction, data needs to be collected to train the trajectory prediction network.
However, in the current trajectory prediction of automatic driving, the trajectory prediction network obtained by training has obvious performance gap on a training data set and a testing data set, so that the generalization capability of the trajectory prediction network is insufficient, and the effect of performing the trajectory prediction is poor.
Disclosure of Invention
The present application is proposed to solve the above-mentioned technical problems. Embodiments of the present application provide a training data acquisition method, an apparatus, a computer-readable storage medium, and an electronic device, which can effectively increase training data in a training data set and improve generalization capability of a trajectory prediction network.
According to a first aspect of the present application, there is provided a training data acquisition method, including:
acquiring an original track data set, wherein the original track data set comprises at least one piece of original track data;
mapping the original trajectory data to an occupation grid map, and acquiring corresponding first data to construct a first data set;
processing the first data set according to a first preset mode to obtain a second data set;
and carrying out weighted summation on the first data set and the second data set to obtain a training data set.
According to a second aspect of the present application, there is provided a training data acquisition apparatus including:
the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring an original track data set, and the original track data set comprises at least one piece of original track data;
the second acquisition module is used for mapping the original trajectory data to an occupation raster map and acquiring corresponding first data to construct a first data set;
the third acquisition module is used for processing the first data set according to a first preset mode to obtain a second data set;
and the training data acquisition module is used for carrying out weighted summation on the first data set and the second data set according to a second preset mode to obtain a training data set.
According to a third aspect of the present application, there is provided a computer-readable storage medium storing a computer program for executing the above-described training data acquisition method.
According to a fourth aspect of the present application, there is provided an electronic apparatus comprising:
a processor;
a memory for storing the processor-executable instructions;
the processor is used for reading the executable instruction from the memory and executing the instruction to realize the training data acquisition method.
Compared with the prior art, the training data acquisition method provided by the application has the beneficial effects that: according to the method and the device, the original track data set is processed to obtain the first data set, the second data set is obtained according to the first data set, the new training data set is obtained by adopting a mode of weighting and summing the second data set and the first data set, the difficulty degree of the training data in the obtained training data set is greatly improved compared with that of the first data, so that the data quality of the training data set is integrally improved, and the training efficiency of the track prediction network adopting the training data set and the generalization capability of the track prediction network are improved.
Drawings
The above and other objects, features and advantages of the present application will become more apparent by describing in more detail embodiments of the present application with reference to the attached drawings. The accompanying drawings are included to provide a further understanding of the embodiments of the application and are incorporated in and constitute a part of this specification, illustrate embodiments of the application and together with the description serve to explain the principles of the application. In the drawings, like reference numbers generally represent like parts or steps.
Fig. 1 is a first flowchart illustrating a training data obtaining method according to an exemplary embodiment of the present application.
Fig. 2 is a schematic flowchart of a second data set obtained in a training data acquisition method according to another exemplary embodiment of the present application.
Fig. 3 is a first flowchart illustrating a training data set obtained in a training data obtaining method according to another exemplary embodiment of the present application.
Fig. 4 is a schematic flowchart of a second process for obtaining a training data set in a training data obtaining method according to another exemplary embodiment of the present application.
Fig. 5 is a third schematic flowchart illustrating a training data set obtained in a training data obtaining method according to another exemplary embodiment of the present application.
Fig. 6 is a flowchart illustrating a training data obtaining method according to an exemplary embodiment of the present application.
Fig. 7 is a schematic diagram of a first weighted summation manner in a training data acquisition method according to another exemplary embodiment of the present application.
Fig. 8 is a schematic diagram of a second weighted summation manner in a training data acquisition method according to another exemplary embodiment of the present application.
Fig. 9 is a schematic diagram of a third weighted summation manner in a training data acquisition method according to another exemplary embodiment of the present application.
Fig. 10 is a first schematic diagram of a training data acquiring apparatus according to an exemplary embodiment of the present application.
Fig. 11 is a schematic diagram of a third obtaining module in the training data obtaining apparatus according to an exemplary embodiment of the present application.
Fig. 12 is a first schematic diagram of a training data acquisition module in a training data acquisition apparatus according to an exemplary embodiment of the present application.
Fig. 13 is a second schematic diagram of a training data obtaining module in a training data obtaining apparatus according to an exemplary embodiment of the present application.
Fig. 14 is a third schematic diagram of a training data acquisition module in the training data acquisition apparatus according to an exemplary embodiment of the present application.
Fig. 15 is a schematic diagram of a training data acquiring apparatus according to an exemplary embodiment of the present application.
Fig. 16 is a block diagram of an electronic device provided in an exemplary embodiment of the present application.
Detailed Description
Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be understood that the described embodiments are only some embodiments of the present application and not all embodiments of the present application, and that the present application is not limited by the example embodiments described herein.
Summary of the application
In the field of autonomous driving, trajectory prediction of obstacles or vehicles is very important for an autonomous driving system to perform path planning, so as to ensure that a path planning algorithm can handle responsible road conditions. In order to better perform the trajectory prediction, the trajectory data needs to be collected to train the trajectory prediction network. At present, when a trajectory prediction network obtained through training of a training data set is tested through test data, a very obvious performance gap usually exists, so that the trajectory prediction effect of the trajectory prediction network in practical application is poor, and the generalization capability of the trajectory prediction network is limited.
In order to improve the generalization capability of the trajectory prediction network, data enhancement needs to be performed on the training data to expand the number of the training data, and the expansion mode of the training data includes mirror inversion, random cutting, random rotation and the like performed on the training data. However, although the amount of training data can be expanded in this way, the expanded training data is very similar to the original training data, so the added training information is very limited, resulting in low training efficiency and limited generalization capability of the trajectory prediction network obtained by training.
The embodiment provides a brand-new method for acquiring the training data, which can improve the quality of the training data while expanding the quantity of the training data, and can effectively improve the generalization capability of the trajectory prediction network when the acquired training data is used for training the trajectory prediction network.
Exemplary method
Fig. 1 is a schematic flowchart of a training data acquisition method according to an exemplary embodiment of the present application. The embodiment can be applied to a server or an electronic device, and is not limited herein. As shown in fig. 1, the training data acquisition method includes the following steps:
step 10: an original trajectory data set is obtained, the original trajectory data set including at least one piece of original trajectory data.
In this embodiment, the raw trajectory data set may include a moving trajectory of the vehicle, and images may be collected by the lidar at a preset frequency, where the collected images include the moving trajectory of the vehicle. By identifying the images, the original trajectory data therein is acquired, so that an original trajectory data set can be obtained. It is understood that the track data in each frame of image is a piece of original track data. Of course, in other embodiments, the raw trajectory data may be obtained in other ways, and is not limited to the above situation, and is not limited herein.
Step 20: and mapping the original trajectory data to an occupation grid map, and acquiring corresponding first data to construct a first data set.
An occupied grid map is a common map representation method, and refers to an image which is discretized in both space and brightness, and the image is divided into a series of grids, each grid is given a possible value to represent the probability that the grid is occupied, so that the positions in the map are reflected, and the positions are not occupied.
After the raw trajectory data is obtained, the digitized raw trajectory data may be mapped to obtain corresponding occupancy grid maps, where each occupancy grid map is a first data. When mapping is performed, only a part of original track data in the original track data set may be mapped, or all of the original track data may be mapped, and the first data obtained after mapping may form the first data set.
Step 30: and processing the first data set according to a first preset mode to obtain a second data set.
In order to expand the data set, the first data in the first data set may be processed, and the specific manner of processing may be selected as needed. Corresponding second data can be obtained through processing, and thus a second data set can be obtained.
Step 40: and carrying out weighted summation on the first data set and the second data set to obtain a training data set.
The second data set obtained through step 30 includes at least one piece of second data, and the quantity of the second data is the same as that of the first data. By performing weighted summation on the first data and the second data, new training data can be obtained, and the weighted weights and the specific manner of weighting can be set as required, which is not limited herein. It will be appreciated that the obtained training data set may include not only the newly obtained training data, but also the first data in the original first data set. The difficulty degree of new training data obtained by weighting and summing the data in the first data set and the second data set is greatly improved compared with that of the first data, so that when the part of training data is used for training the trajectory prediction network, the training efficiency of the trajectory prediction network can be effectively improved.
The training data acquisition method provided by the embodiment has the beneficial effects that: in the embodiment, the original trajectory data set is processed to obtain the first data set, the second data set is obtained according to the first data set, and the new training data set is obtained by weighting and summing the second data set and the first data set, so that the difficulty degree of the training data in the obtained training data set is greatly improved compared with that of the first data, the data quality of the training data set is integrally improved, and the training efficiency of the trajectory prediction network adopting the training data set and the generalization capability of the trajectory prediction network are improved.
Fig. 2 is a schematic flow chart of processing the first data set according to a first preset mode to obtain a second data set in the training data obtaining method according to an exemplary embodiment of the present application. As shown in fig. 2, step 30 may include the steps of:
step 301: and copying the first data in the first data set to obtain an intermediate data set.
Step 302: randomly arranging the data in the intermediate data set to obtain the second data set.
In this embodiment, the first data in the first data set may be arranged in the order of obtaining the image frames, and may be obtained by copying the first data in the first data set when obtaining the second data set. In order to increase the difficulty of subsequently obtained training data, the order of the data in the obtained intermediate data set may be rearranged such that the order of the second data in the obtained second data set is different from the order of the first data in the first data set, so as to facilitate subsequent processing of the data in the two data sets. For example, the first data set includes first sub data, second sub data, third sub data, and fourth sub data (all being first data, written as sub data here for distinction) arranged in order, and the second data set obtained by copying and randomly arranging the data may change the arrangement order of the data into: the second sub-data, the third sub-data, the fourth sub-data and the first sub-data.
In the embodiment, when the second data set is constructed, the first data in the first data set is copied, and the arrangement mode of the first data is disturbed for rearrangement, so that the obtained second data set is different from the first data set, which is beneficial to improving the difficulty degree of the training data obtained by the first data set and the second data set.
In this embodiment, in the step of performing weighted summation on the first data set and the second data set to obtain the training data set, a specific manner of performing weighted summation on the first data set and the second data set may be set as required.
For example, a first way of weighted summation is shown in fig. 3, and step 40 may include the steps of:
step 401: and performing space division on the data in the first data set and the data in the second data set according to a second preset mode, so that the data in the first data set and the data in the second data set are divided into at least two subspaces.
In this embodiment, the first data in the first data set is an occupancy grid map corresponding to the original trajectory data, and each first data includes a plurality of data points scattered on a two-dimensional spatial plane. The second preset mode can be equal division or unequal division of the two-dimensional space plane of the first data; the two-dimensional space plane can be divided into a plurality of square areas, and can also be divided into a plurality of annular rings from the center of the plane space to the outside. Of course, the division into spaces may be performed in other ways, and is not limited to the above-described case.
Referring to fig. 7, the embodiment is described by taking the example of dividing the space into 4 subspaces. For processing the first data in the first data set, the space of the first data may be equally divided into 4 subspaces of 2 × 2, and for convenience of description, the four subspaces of the first data are respectively the first subspace 611, the second subspace 612, the third subspace 613, and the fourth subspace 614 (as shown in fig. 7 a). The data points in each subspace correspond to the data points divided into that subspace. It will be appreciated that when the subspace partitioning is performed, all of the first data in the first data set is similarly partitioned. Similarly, the four subspaces of the second data are the fifth subspace 621, the sixth subspace 622, the seventh subspace 623, and the eighth subspace 624, respectively (as shown in fig. 7 b). The data points in each subspace correspond to the data points divided into that subspace. It will be appreciated that when the subspace partitioning is performed, all of the second data in the second data set is similarly partitioned. The first subspace 611 corresponds to the fifth subspace 621, the second subspace 612 corresponds to the sixth subspace 622, the third subspace 613 corresponds to the seventh subspace 623, and the fourth subspace 614 corresponds to the eighth subspace 624.
Step 402: and processing the data of different subspaces of the first data set by adopting a first coefficient to correspondingly obtain a first spatial data set.
The number of the first coefficients corresponds to the number of the subspaces divided by the first data set, i.e. when the number of the subspaces is 4, the number of the first coefficients is also 4, and for convenience of description, the first coefficients are respectively denoted as λ11、λ12、λ13、λ14. And multiplying the coordinates of the data points of different subspaces by corresponding coefficients respectively to obtain new data points, wherein the new data points form a first spatial data set.
Specifically, let the coordinate of the data point of the first subspace 611 in the first data set be (x)11,y11) The processed coordinates of the data points of the first subspace 611 are (λ)11x11,λ11y11) (ii) a The coordinates of the data points of the second subspace 612 are (x)12,y12) The processed coordinates of the data points of the second subspace 612 are (λ)12x12,λ12y12) (ii) a The coordinates of the data point of the third subspace 613 are (x)13,y13) The processed coordinates of the data points of the third subspace 613 are (λ)13x13,λ13y13) (ii) a The coordinates of the data points of the fourth subspace 614 are (x)14,y14) The processed coordinates of the data points of the fourth subspace 614 are (λ)14x14,λ14y14)。
Step 403: and processing the data of different subspaces of the second data set by adopting a second coefficient to correspondingly obtain a second spatial data set.
The number of the second coefficients corresponds to the number of the subspaces divided by the second data set, i.e. when the number of the subspaces is 4, the number of the first coefficients is also 4, and for convenience of description, the second coefficients are respectively denoted as λ21、λ22、λ23、λ24. And multiplying the coordinates of the data points of different subspaces by corresponding coefficients respectively to obtain new data points, wherein the new data points form a second spatial data set.
Specifically, let the coordinate of the data point in the fifth subspace 621 in the second data set be (x)21,y21) The processed coordinates of the data points in the second subspace 621 are (λ)21x21,λ21y21) (ii) a The coordinates of the data points of the sixth subspace 622 are (x)22,y22) The processed coordinates of the data points in the sixth subspace 622 are (λ)22x22,λ22y22) (ii) a The coordinates of the data points of the seventh subspace 623 are (x)23,y23) The processed coordinates of the data points in the seventh subspace 623 are (λ)23x23,λ23y23) (ii) a The coordinates of the data points of the eighth subspace 624 are (x)24,y24) The processed coordinates of the data points in the eighth subspace 624 are (λ)24x24,λ24y24)。
Step 404: and correspondingly weighting and summing the first spatial data set and the second spatial data set to obtain the training data set.
Specifically, the training data set includes a plurality of training data, and each training data may be obtained by weighted summation of corresponding first data and second data. For each data point in the training data, the subspace of the corresponding first data is different according to the position. Here, the data point correspondence in the training data is denoted as (x)31,y31)、(x32,y32)、(x33,y33)、(x34,y34) Wherein x is31＝λ11x11+λ21x21，y31＝λ11y11+λ21y21；x32＝λ12x12+λ22x22，y32＝λ12y12+λ22y22；x33＝λ13x13+λ23x23，y33＝λ13y13+λ23y23；x34＝λ14x14+λ24x24，y34＝λ14y14+λ24y24。
According to the embodiment, the space of the first data set and the space of the second data set are divided, different subspaces are multiplied by different coefficients to be processed, and then the data of different subspaces are weighted and summed, so that new training data can be obtained, the obtained new training data has a large difference with the first data, the number and diversity of the training data can be effectively expanded, the difficulty degree of the training data is improved, and the efficiency of subsequent trajectory prediction network training is facilitated.
Further, the obtaining mode of the first coefficient and the second coefficient can be selected according to the requirement. For example, the first coefficient may be randomly chosen from a beta function distribution (beta distribution, i.e., a set of continuous probability distributions defined in the (0,1) interval). The sum of the second coefficient and the corresponding first coefficient is a fixed value, and the fixed value may be 1 or another value. Let us say that the value is 1, λ11+λ21＝1，λ12+λ22＝1，λ13+λ23＝1，λ14+λ241. Of course, the fixed value may be other values, and is not limited herein. In this embodiment, the first coefficient and the second coefficient are selected in the above manner, and the linear difference is performed on the first data in the data mixing manner, so as to construct a larger and denser training data set distributed in the current training data set.
As another example, a second way of weighted summation is shown in fig. 4, and step 40 may include the steps of:
step 411: and dividing the first data set and the second data set according to a third preset mode to correspondingly obtain at least two first sub data sets and at least two second sub data sets.
In this embodiment, when the data set is divided by using the third preset manner, the data set is divided into a plurality of sub data sets instead of dividing the space of the data set, and the number of the sub data sets may be set as needed. Referring to fig. 8, for example, the number of the first data in the first data set is 4, the 4 first data are divided into 4 first sub data sets, which are respectively recorded as a first sub data set, a second sub data set, a third sub data set and a fourth sub data set (as shown in fig. 8 a); correspondingly, the number of the second data in the second data set is 4, and the 4 second data sets are divided into 4 second sub data sets, which are respectively denoted as sub data set five, sub data set six, sub data set seven, and sub data set eight (as shown in fig. 8 b).
Step 412: and processing the different first sub data set and the second sub data set by adopting different coefficients to correspondingly obtain a first sub intermediate data set and a second sub intermediate data set.
In this embodiment, different first coefficients may be used to process the 4 first sub-data sets, for example, the number of the first coefficients is four, which is respectively denoted as λ11、λ12、λ13、λ14(as shown in fig. 8 a). The 4 second sub data sets are processed by using different second coefficients, for example, the number of the second coefficients is four, and the second coefficients are respectively marked as λ21、λ22、λ23、λ24(as shown in fig. 8 b).
For convenience of description, the data point of the first data set is a subset of the first data set having coordinates (x)11,y11) The processed coordinate of the data point of the subdata set one is (lambda)11x11,λ11y11) (ii) a The coordinates of the data point of the second subset of data in the first data set are (x)12,y12) The processed coordinate of the data point of the sub data set two is (lambda)12x12,λ12y12) (ii) a The coordinate of the data point of the third subset of data in the first data set is (x)13,y13) The processed coordinate of the data point of the subdata set three is (lambda)13x13,λ13y13) (ii) a The coordinate of the data point of sub data set four in the first data set is (x)14,y14) The processed coordinate of the data point of the subdata set four is (lambda)14x14,λ14y14)。
Similarly, the data point of sub data set five in the second data set has coordinates of (x)21,y21) The processed coordinate of the data point of the sub data set five is (lambda)21x21,λ21y21) (ii) a The coordinates of the data point of sub-data set six in the second data set are (x)22,y22) The processed coordinate of the data point of the sub data set six is (lambda)22x22,λ22y22) (ii) a The coordinates of the data point of the second data set subset seven are (x)23,y23) Data of, data ofThe processed coordinate of the data point of the set seven is (lambda)23x23,λ23y23) (ii) a The data point of the subset eight of the second data set has the coordinate of (x)24,y24) The processed coordinate of the data point of the sub data set eight is (lambda)24x24,λ24y24)。
Step 413: and correspondingly weighting and summing the first sub-intermediate data set and the second sub-intermediate data set to obtain the training data set.
Specifically, the training data set includes a plurality of training data, and each training data may be obtained by weighted summation of corresponding first data and second data. For the data points in the training data set, the sub data sets of the corresponding first data are different according to different training data. Here, the data point correspondences in the different training data are denoted as (x)31,y31)、(x32,y32)、(x33,y33)、(x34,y34) Wherein x is31＝λ11x11+λ21x21，y31＝λ11y11+λ21y21；x32＝λ12x12+λ22x22，y32＝λ12y12+λ22y22；x33＝λ13x13+λ23x23，y33＝λ13y13+λ23y23；x34＝λ14x14+λ24x24，y34＝λ14y14+λ24y24。
According to the embodiment, the first data set and the second data set are divided, different sub data sets are processed by different coefficients, and then the data of the different sub data sets are subjected to weighted summation, so that new training data can be obtained, the obtained new training data has a large difference with the first data, the number and diversity of the training data can be effectively expanded, the difficulty degree of the training data is improved, and the efficiency of subsequent trajectory prediction network training is facilitated.
As another example, a third way of weighted summation is shown in fig. 5, and step 40 may include the following steps:
step 421: the first data set is processed with first coefficients to obtain a first intermediate data set.
Referring to fig. 9, in the method, when the first data set is processed, the first data set is not distinguished, but all data in the first data set are processed by using the same first coefficient (as shown in fig. 9 a). For the convenience of description, the first coefficient is denoted as λ1The data points in the first data set are denoted as (x)1,y1) Then the data point in the first intermediate dataset is obtained as (λ)1x1,λ1y1)。
Step 422: the second data set is processed with second coefficients to obtain a second intermediate data set.
Referring to fig. 9, similarly, in the method, when the second data set is processed, the second data set is not distinguished, but all data in the second data set are processed by using the same second coefficient (as shown in fig. 9 b). For the convenience of description, the second coefficient is denoted as λ2The data points in the first data set are denoted as (x)2,y2) Then the data point in the first intermediate dataset is obtained as (λ)2x2,λ2y2)。
Step 423: and carrying out weighted summation on the first intermediate data set and the second intermediate data set to obtain the training data set.
Specifically, the training data set includes a plurality of training data, and each training data may be obtained by weighted summation of corresponding first data and second data. The data points in the training data set are denoted as (x)3,y3) Wherein x is3＝λ1x1+λ2x2，y3＝λ1y1+λ2y2. The obtaining mode of the first coefficient and the second coefficient can be selected according to requirements. For example, the firstA coefficient may be randomly chosen from a beta distribution. The sum of the second coefficient and the corresponding first coefficient is a fixed value, and the fixed value may be 1 or another value. Let us say that the value is 1, λ1+λ2＝1。
The embodiment respectively processes the first data set and the second data set through the first coefficient and the second coefficient, and then performs weighted summation, so that new training data can be obtained, the operation is simple, the obtained new training data has a large difference with the first data, the quantity and diversity of the training data can be effectively expanded, the difficulty degree of the training data is improved, and the efficiency of subsequent trajectory prediction network training is facilitated.
Further, after the training data is obtained, the training data may also be applied. Fig. 6 is a flowchart illustrating a training data obtaining method according to an exemplary embodiment of the present application. As shown in fig. 6, the following steps may be further included after step 40:
step 50: and inputting the training data set into a track prediction network so as to train the track prediction network.
The trajectory prediction network is used to predict a trajectory, for example, a trajectory of a vehicle may be predicted. Before the trajectory prediction network is used for trajectory prediction, it needs to be trained to ensure good performance. The training data set provided by the embodiment not only expands the training data, but also increases the number of difficult samples in the process of expanding the training data, so that the training speed of the trajectory prediction network can be accelerated, the training efficiency of the trajectory prediction network is effectively improved, the generalization capability of the trajectory prediction network is further improved, and the trajectory prediction network has a good prediction effect when performing trajectory prediction.
Exemplary devices
Fig. 10 is a schematic diagram of a training data acquisition apparatus according to an exemplary embodiment of the present application. As shown in fig. 10, the training data acquisition means includes a first acquisition module 71, a second acquisition module 72, a third acquisition module 73, and a training data acquisition module 74. The first obtaining module 71 is configured to obtain an original trajectory data set, where the original trajectory data set includes at least one piece of original trajectory data; the second obtaining module 72 is configured to map the raw trajectory data to an occupancy grid map, and obtain corresponding first data to construct a first data set; the third obtaining module 73 is configured to process the first data set according to a first preset manner, so as to obtain a second data set; the training data obtaining module 74 is configured to perform weighted summation on the first data set and the second data set according to a second preset manner, so as to obtain a training data set.
Referring to fig. 11, further, the third obtaining module 73 includes a copying unit 731 and an arranging unit 732. The copying unit 731 is configured to copy first data in the first data set to obtain an intermediate data set; the permutation unit 732 is configured to randomly permute the data in the intermediate data set to obtain the second data set.
Referring to fig. 12, in one embodiment, the training data obtaining module 74 includes a space dividing unit 740, a first data processing unit 741, a second data processing unit 742, and a first data obtaining unit 743. The space dividing unit 740 is configured to perform space division on the data in the first data set and the data in the second data set according to a second preset manner, so that the data in the first data set and the data in the second data set are divided into at least two subspaces; the first data processing unit 741 is configured to process data of different subspaces of the first data set by using a first coefficient, and correspondingly obtain a first spatial data set; the second data processing unit 742 is configured to process data of different subspaces of the second data set by using a second coefficient, and correspondingly obtain a second spatial data set; the first data obtaining unit 743 is configured to perform a weighted summation on the first spatial data set and the second spatial data set, so as to obtain the training data set.
Referring to fig. 13, in one embodiment, the training data obtaining module 74 includes a data set dividing unit 744, a third data processing unit 745, and a second data obtaining unit 746. The data set dividing unit 744 is configured to divide the first data set and the second data set according to a third preset manner, and correspondingly obtain at least two first sub data sets and at least two second sub data sets; the third data processing unit 745 is configured to process the different first sub data set and the second sub data set by using different coefficients, so as to obtain a first sub intermediate data set and a second sub intermediate data set correspondingly; the second data obtaining unit 746 is configured to sum the first sub-intermediate data set and the second sub-intermediate data set in a weighted manner, so as to obtain the training data set.
Referring to fig. 14, in one embodiment, the training data obtaining module 74 includes a fourth data processing unit 747, a fifth data processing unit 748 and a third data obtaining unit 749. Wherein the fourth data processing unit 747 is configured to process the first data set with the first coefficients to obtain a first intermediate data set; the fifth data processing unit 748 is configured to process the second data set with second coefficients to obtain a second intermediate data set; the third data obtaining unit 749 is configured to perform weighted summation on the first intermediate data set and the second intermediate data set to obtain the training data set.
Referring to fig. 15, further, the training data obtaining apparatus further includes a training module 75, and the training module 75 is configured to input the training data set into the trajectory prediction network to train the trajectory prediction network.
Exemplary electronic device
FIG. 16 illustrates a block diagram of an electronic device in accordance with an embodiment of the present application.
As shown in fig. 16, the electronic device 8 includes one or more processors 81 and memory 82.
The processor 81 may be a Central Processing Unit (CPU) or other form of processing unit having data processing capabilities and/or instruction execution capabilities, and may control other components in the electronic device 8 to perform desired functions.
Memory 82 may include one or more computer program products that may include various forms of computer-readable storage media, such as volatile memory and/or non-volatile memory. The volatile memory may include, for example, Random Access Memory (RAM), cache memory (cache), and/or the like. The non-volatile memory may include, for example, Read Only Memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium and executed by the processor 81 to implement the training data acquisition methods of the various embodiments of the present application described above and/or other desired functions.
In one example, the electronic device 8 may further include: an input device 83 and an output device 84, which are interconnected by a bus system and/or other form of connection mechanism (not shown). The input means 83 may be a communication network connector when the electronic device is a stand-alone device. The input device 83 may include, for example, a keyboard, a mouse, and the like. The output device 84 may output various information to the outside, and may include, for example, a display, speakers, a printer, and a communication network and its connected remote output devices, among others.
Of course, for the sake of simplicity, only some of the components related to the present application in the electronic device 8 are shown in fig. 16, and components such as a bus, an input/output interface, and the like are omitted. In addition, the electronic device 8 may include any other suitable components, depending on the particular application.
Exemplary computer program product and computer-readable storage Medium
In addition to the above-described methods and devices, embodiments of the present application may also be a computer program product comprising computer program instructions which, when executed by a processor, cause the processor to perform the steps in the sound source localization method according to various embodiments of the present application described in the above-mentioned "exemplary methods" section of the present description.
The computer program product may be written with program code for performing the operations of embodiments of the present application in any combination of one or more programming languages, including an object oriented programming language such as Java, C + + or the like and conventional procedural programming languages, such as the "C" programming language or similar programming languages. The program code may execute entirely on the user's computing device, partly on the user's device, as a stand-alone software package, partly on the user's computing device and partly on a remote computing device, or entirely on the remote computing device or server.
Furthermore, embodiments of the present application may also be a computer-readable storage medium having stored thereon computer program instructions that, when executed by a processor, cause the processor to perform the steps in the training data acquisition method according to various embodiments of the present application described in the "exemplary methods" section above in this specification.
The computer-readable storage medium may take any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. A readable storage medium may include, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or a combination of any of the foregoing. More specific examples (a non-exhaustive list) of the readable storage medium include: an electrical connection having one or more wires, a portable disk, a hard disk, a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.
The foregoing describes the general principles of the present application in conjunction with specific embodiments, however, it is noted that the advantages, effects, etc. mentioned in the present application are merely examples and are not limiting, and they should not be considered essential to the various embodiments of the present application. Furthermore, the foregoing disclosure of specific details is for the purpose of illustration and description and is not intended to be limiting, since the foregoing disclosure is not intended to be exhaustive or to limit the disclosure to the precise details disclosed.
The block diagrams of devices, apparatuses, systems referred to in this application are only given as illustrative examples and are not intended to require or imply that the connections, arrangements, configurations, etc. must be made in the manner shown in the block diagrams. These devices, apparatuses, devices, systems may be connected, arranged, configured in any manner, as will be appreciated by those skilled in the art. Words such as "including," "comprising," "having," and the like are open-ended words that mean "including, but not limited to," and are used interchangeably therewith. The words "or" and "as used herein mean, and are used interchangeably with, the word" and/or, "unless the context clearly dictates otherwise. The word "such as" is used herein to mean, and is used interchangeably with, the phrase "such as but not limited to".
It should also be noted that in the devices, apparatuses, and methods of the present application, the components or steps may be decomposed and/or recombined. These decompositions and/or recombinations are to be considered as equivalents of the present application.
The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.
The foregoing description has been presented for purposes of illustration and description. Furthermore, the description is not intended to limit embodiments of the application to the form disclosed herein. While a number of example aspects and embodiments have been discussed above, those of skill in the art will recognize certain variations, modifications, alterations, additions and sub-combinations thereof.

Claims (10)

1. A training data acquisition method, comprising:
acquiring an original track data set, wherein the original track data set comprises at least one piece of original track data;
mapping the original trajectory data to an occupation grid map, and acquiring corresponding first data to construct a first data set;
processing the first data set according to a first preset mode to obtain a second data set;
and carrying out weighted summation on the first data set and the second data set to obtain a training data set.
2. The method of claim 1, wherein the processing the first data set according to a first preset manner to obtain a second data set comprises:
copying first data in the first data set to obtain an intermediate data set;
randomly arranging the data in the intermediate data set to obtain the second data set.
3. The method of claim 1, wherein the weighted summation of the first data set and the second data set to obtain a training data set comprises:
performing spatial division on data in the first data set and the second data set according to a second preset mode, so that the data in the first data set and the second data set are divided into at least two subspaces;
processing data of different subspaces of the first data set by adopting a first coefficient to correspondingly obtain a first spatial data set;
processing the data of different subspaces of the second data set by adopting a second coefficient to correspondingly obtain a second spatial data set;
and correspondingly weighting and summing the first spatial data set and the second spatial data set to obtain the training data set.
4. The method of claim 3, wherein the first coefficients are randomly selected from a beta function distribution, and the number of the first coefficients corresponds to the number of subspaces of the first data set;
the number of the second coefficients is the same as that of the first coefficients, and the sum of the first coefficients and the corresponding second coefficients is a fixed value.
5. The method of claim 1, wherein the weighted summation of the first data set and the second data set to obtain a training data set comprises:
dividing the first data set and the second data set according to a third preset mode to correspondingly obtain at least two first sub data sets and at least two second sub data sets;
processing the different first sub data set and the second sub data set by adopting different coefficients to correspondingly obtain a first sub intermediate data set and a second sub intermediate data set;
and correspondingly weighting and summing the first sub-intermediate data set and the second sub-intermediate data set to obtain the training data set.
6. The method of claim 1, wherein the weighted summation of the first data set and the second data set to obtain a training data set comprises:
processing the first data set with first coefficients to obtain a first intermediate data set;
processing the second data set with a second coefficient to obtain a second intermediate data set;
and carrying out weighted summation on the first intermediate data set and the second intermediate data set to obtain the training data set.
7. The method according to any one of claims 1-6, wherein after the step of performing a weighted summation on the first data set and the second data set according to a second preset manner to obtain a training data set, the method further comprises:
and inputting the training data set into a track prediction network so as to train the track prediction network.
8. A training data acquisition apparatus comprising:
the system comprises a first acquisition module, a second acquisition module and a third acquisition module, wherein the first acquisition module is used for acquiring an original track data set, and the original track data set comprises at least one piece of original track data;
the second acquisition module is used for mapping the original trajectory data to an occupation raster map and acquiring corresponding first data to construct a first data set;
the third acquisition module is used for processing the first data set according to a first preset mode to obtain a second data set;
and the training data acquisition module is used for carrying out weighted summation on the first data set and the second data set according to a second preset mode to obtain a training data set.
9. A computer-readable storage medium storing a computer program for executing the training data acquisition method according to any one of claims 1 to 7.
10. An electronic device, the electronic device comprising:
a processor;
a memory for storing the processor-executable instructions;
the processor is configured to read the executable instructions from the memory and execute the instructions to implement the training data obtaining method of any one of claims 1 to 7.
CN202010038695.8A 2020-01-14 2020-01-14 Training data acquisition method and device Pending CN113191173A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202010038695.8A CN113191173A (en) 2020-01-14 2020-01-14 Training data acquisition method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202010038695.8A CN113191173A (en) 2020-01-14 2020-01-14 Training data acquisition method and device

Publications (1)

Publication Number Publication Date
CN113191173A true CN113191173A (en) 2021-07-30

Family

ID=76972406

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202010038695.8A Pending CN113191173A (en) 2020-01-14 2020-01-14 Training data acquisition method and device

Country Status (1)

Country Link
CN (1) CN113191173A (en)

Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130085344A1 (en) * 2011-10-04 2013-04-04 Medtronic Navigation, Inc. Method and Apparatus for Assisted Trajectory Planning
CN108764372A (en) * 2018-06-08 2018-11-06 Oppo广东移动通信有限公司 Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set
CN109885378A (en) * 2019-01-04 2019-06-14 平安科技（深圳）有限公司 Model training method, device, computer equipment and computer readable storage medium
CN110414732A (en) * 2019-07-23 2019-11-05 中国科学院地理科学与资源研究所 A kind of trip Future Trajectory Prediction method, apparatus, storage medium and electronic equipment

Patent Citations (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20130085344A1 (en) * 2011-10-04 2013-04-04 Medtronic Navigation, Inc. Method and Apparatus for Assisted Trajectory Planning
CN108764372A (en) * 2018-06-08 2018-11-06 Oppo广东移动通信有限公司 Construction method and device, mobile terminal, the readable storage medium storing program for executing of data set
CN109885378A (en) * 2019-01-04 2019-06-14 平安科技（深圳）有限公司 Model training method, device, computer equipment and computer readable storage medium
CN110414732A (en) * 2019-07-23 2019-11-05 中国科学院地理科学与资源研究所 A kind of trip Future Trajectory Prediction method, apparatus, storage medium and electronic equipment

Non-Patent Citations (1)

* Cited by examiner, † Cited by third party
Title
张金旺;章永进;徐友春;: "基于概率统计的车辆运动轨迹预测方法", 军事交通学院学报, no. 08 *

Similar Documents

Publication Publication Date Title
JP6441980B2 (en) Method, computer and program for generating teacher images
EP3792830A1 (en) Training a class-conditional generative adverserial network
US11816559B2 (en) Dilated convolution using systolic array
JP2020042822A (en) Ground detection method, apparatus, electronic device, vehicle, and storage medium
US20160239597A1 (en) Apparatus and method for performing finite element computation
US11144291B1 (en) Loop-oriented neural network compilation
EP4020317A1 (en) Methods and apparatus for high-fidelity vision tasks using deep neural networks
US20200110988A1 (en) Computing device and method
WO2005008596A2 (en) Method, computer program and computer readable means for projecting data from a multidimensional space into a space having less dimensions and to carry out a cognitive analysis on said data
WO2019224164A1 (en) Methods, systems, articles of manufacture and apparatus to reconstruct scenes using convolutional neural networks
US11501145B1 (en) Memory operation for systolic array
CN110633735B (en) Progressive depth convolution network image identification method and device based on wavelet transformation
JP2020529060A (en) Prediction of molecular properties of molecular variants using residue-specific molecular structural features
CN113920538A (en) Object detection method, device, equipment, storage medium and computer program product
CN116261734A (en) Neural architecture scaling for hardware accelerators
CN113191173A (en) Training data acquisition method and device
CN114631099A (en) Artificial intelligence transparency
WO2023136020A1 (en) Pathfinding apparatus, pathfinding method, and non-transitory computer-readable storage medium
Zhang et al. Online task scheduling for LiDAR data preprocessing on hybrid GPU/CPU devices: A reinforcement learning approach
EP3968202A1 (en) Customizable reinforcement of learning column placement in structural design
US20220036191A1 (en) System and Related Methods for Reducing the Resource Consumption of a Convolutional Neural Network
CN115761599A (en) Video anomaly detection method and system
JPWO2020054402A1 (en) Neural network processing device, computer program, neural network manufacturing method, neural network data manufacturing method, neural network utilization device, and neural network miniaturization method
US20220379919A1 (en) Parameter space optimization
JP2004527172A (en) Scalable system and method for optimizing a random system of algorithms for image quality

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination