Search

CN-121997268-A - Data labeling method, device, vehicle, storage medium and program product

CN121997268ACN 121997268 ACN121997268 ACN 121997268ACN-121997268-A

Abstract

The embodiment of the application provides a method, a device, a vehicle, a storage medium and a program product for labeling data, wherein the method comprises the steps of acquiring modal data under at least one mode acquired by a plurality of sensors deployed in the vehicle to obtain multi-modal data; the method comprises the steps of inputting multi-mode data into a labeling model to carry out labeling, and obtaining at least one initial labeling result, wherein the labeling model is obtained by training a multi-mode data sample acquired based on a sensor and a labeling result sample corresponding to the multi-mode data sample, the initial labeling result is at least used for representing the position of at least one obstacle in a traffic area, responding to the fact that the confidence of the initial labeling result is lower than a confidence threshold value, obtaining an adjustment instruction, responding to the adjustment instruction, and adjusting three-dimensional space labels of the obstacle in the initial labeling result to obtain a target labeling result. The application solves the technical problem of low accuracy of data processing.

Inventors

  • GAO TIANYI

Assignees

  • 奇瑞汽车股份有限公司

Dates

Publication Date
20260508
Application Date
20260130

Claims (10)

  1. 1. A method for labeling data, comprising: Acquiring modal data under at least one mode acquired by a plurality of sensors deployed in a vehicle to obtain multi-modal data, wherein the multi-modal data is used for representing traffic states when the vehicle is located in a traffic area, and the traffic area is located in a target area range; Inputting the multi-mode data into a labeling model to carry out labeling to obtain at least one initial labeling result, wherein the labeling model is obtained by training based on a multi-mode data sample acquired by the sensor and a labeling result sample corresponding to the multi-mode data sample, and the initial labeling result is at least used for representing the position of at least one obstacle in the traffic area; acquiring an adjustment instruction in response to the confidence level of the initial labeling result being lower than a confidence level threshold; And responding to the adjustment instruction, and adjusting the three-dimensional space label of the obstacle in the initial labeling result to obtain a target labeling result, wherein the three-dimensional space label is used for representing the position and the size of the obstacle.
  2. 2. The method of claim 1, wherein the inputting the multi-modal data into a labeling model for labeling, to obtain at least one initial labeling result, comprises: According to the time stamp, performing time sequence alignment on the multi-mode data; Correcting the multi-mode data after time sequence alignment to obtain a plurality of correction data; and inputting a plurality of correction data into the labeling model for labeling, and obtaining at least one initial labeling result.
  3. 3. The method according to claim 2, wherein the inputting the plurality of correction data into the labeling model for labeling, to obtain at least one initial labeling result, includes: and inputting prompt information and a plurality of correction data into the marking model to obtain at least one initial marking result, wherein the prompt information is used for prompting the obstacle to be marked to the marking model.
  4. 4. The method of claim 1, wherein the obtaining an adjustment instruction in response to the confidence level of the initial labeling result being below a confidence threshold comprises: marking the initial marking result in at least one mode data corresponding to the initial marking result in response to the confidence coefficient of the initial marking result being lower than the confidence coefficient threshold; Displaying the marked modal data in a display interface; Acquiring operation data in the display interface; And converting the operation data into the adjustment instruction.
  5. 5. The method according to any one of claims 1 to 4, further comprising: Under a plurality of continuous moments, determining the similarity between a plurality of target labeling results corresponding to the obstacle, wherein the target labeling results and the moments are in one-to-one correspondence; In response to the similarity being below a similarity threshold, the obstacle is marked and the obstacle is re-marked.
  6. 6. The method according to any one of claims 1 to 4, further comprising: responding to the confidence level of the initial labeling result being lower than the confidence level threshold, taking the multi-modal data as the multi-modal data sample, taking the initial labeling result as the labeling result sample, and storing the initial labeling result into a database; Responding to the fact that the number of the multi-mode data samples in the database exceeds a data volume threshold, selecting a target number of the multi-mode data samples from the database, and marking result samples corresponding to the multi-mode data samples; and updating the labeling model by using the selected multi-mode data sample and the labeling result sample corresponding to the multi-mode data sample.
  7. 7. A device for labeling data, comprising: The system comprises a first acquisition unit, a second acquisition unit and a control unit, wherein the first acquisition unit is used for acquiring modal data under at least one mode acquired by a plurality of sensors deployed in a vehicle to obtain multi-modal data, the multi-modal data are used for representing traffic states when the vehicle is located in a traffic area, and the traffic area is located in a target area range; The marking unit is used for inputting the multi-mode data into a marking model to carry out marking so as to obtain at least one initial marking result, wherein the marking model is obtained by training based on the multi-mode data sample acquired by the sensor and a marking result sample corresponding to the multi-mode data sample, and the initial marking result is at least used for representing the position of at least one obstacle in the traffic area; The second acquisition unit is used for acquiring an adjustment instruction in response to the confidence coefficient of the initial labeling result being lower than a confidence coefficient threshold value; The adjustment unit is used for responding to the adjustment instruction, adjusting the three-dimensional space label of the obstacle in the initial labeling result to obtain a target labeling result, wherein the three-dimensional space label is used for representing the position and the size of the obstacle.
  8. 8. A vehicle, characterized by comprising: A memory storing an executable program; a processor for running the program, wherein the program runs the method of any one of claims 1 to 6.
  9. 9. A computer readable storage medium, characterized in that the computer readable storage medium comprises a stored executable program, wherein the executable program when run controls a device in which the storage medium is located to run the method according to any one of claims 1 to 6.
  10. 10. A computer program product comprising computer instructions which, when executed by a processor, implement the method of any one of claims 1 to 6.

Description

Data labeling method, device, vehicle, storage medium and program product Technical Field The embodiment of the application relates to the technical field of intelligent driving, in particular to a method and a device for labeling data, a vehicle, a storage medium and a program product. Background Currently, with the development of automatic driving technology, vehicles require a large amount of high-quality multi-modal data, such as image, video, lidar point cloud, semantic tags, and the like, to support the training of perception, decision and control models. In the related art, the data production and the marking are generally completed through manual marking and part of rule-driven automatic marking tools. The manual mode has the advantages of strong operability and high flexibility, but has low labeling efficiency and high cost when facing a large-scale automatic driving data set. The part of automatic marking tools realizes semiautomatic marking by using a rule or a traditional machine learning method, but has limited processing capacity on complex traffic scenes and long-tail data, and still needs a large amount of manual intervention, so the method has the technical problem of low accuracy of data processing. There is currently no good solution to the above problems. Disclosure of Invention The embodiment of the application provides a method, a device, a vehicle, a storage medium and a program product for labeling data, which are used for at least solving the technical problem of low accuracy of data processing. According to one aspect of the embodiment of the application, a method for labeling data is provided, which comprises the steps of obtaining modal data in at least one mode acquired by a plurality of sensors deployed in a vehicle, obtaining multi-modal data, wherein the multi-modal data are used for representing traffic states when the vehicle is located in a traffic area, the traffic area is located in a target area range, inputting the multi-modal data into a labeling model to label, obtaining at least one initial labeling result, wherein the labeling model is obtained by training based on multi-modal data samples acquired by the sensors and labeling result samples corresponding to the multi-modal data samples, the initial labeling result is at least used for representing the position of at least one obstacle in the traffic area, responding to the confidence of the initial labeling result being lower than a confidence threshold, obtaining an adjustment instruction, responding to the adjustment instruction, adjusting three-dimensional space labels of the obstacles in the initial labeling result, and obtaining a target labeling result, wherein the three-dimensional space labels are used for representing the positions and the sizes of the obstacles. The method comprises the steps of obtaining a plurality of initial labeling results, namely, obtaining a plurality of initial labeling results, wherein the initial labeling results comprise the steps of carrying out time sequence alignment on the multi-modal data according to time stamps, correcting the multi-modal data with the time sequence aligned to obtain a plurality of correction data, and inputting the correction data into a labeling model to carry out labeling. Further, inputting a plurality of correction data into the labeling model for labeling to obtain at least one initial labeling result, including inputting prompt information and a plurality of correction data into the labeling model to obtain at least one initial labeling result, wherein the prompt information is used for prompting an obstacle to be labeled to the labeling model. Further, responding to the fact that the confidence coefficient of the initial labeling result is lower than a confidence coefficient threshold value, obtaining an adjustment instruction comprises the steps of labeling the initial labeling result in at least one mode data corresponding to the initial labeling result in response to the fact that the confidence coefficient of the initial labeling result is lower than the confidence coefficient threshold value, displaying the labeled mode data in a display interface, obtaining operation data in the display interface, and converting the operation data into the adjustment instruction. Further, the method can further comprise the steps of determining the similarity between a plurality of target labeling results corresponding to the obstacle at a plurality of continuous moments, wherein the target labeling results correspond to the moments one by one, labeling the obstacle in response to the similarity being lower than a similarity threshold, and labeling the obstacle again. The method further comprises the steps of taking the multi-mode data as a multi-mode data sample and taking the initial labeling result as a labeling result sample and storing the labeling result sample in a database in response to the confidence level of the initial labeling result being lowe