KR-102962562-B1 - Method for generating high efficiency training data on self-driving and apparatus thereof

KR102962562B1KR 102962562 B1KR102962562 B1KR 102962562B1KR-102962562-B1

Abstract

An embodiment of the present invention discloses a method for generating high-efficiency learning data for autonomous driving, comprising: a step of assigning a score to each of a plurality of images; a step of sequentially moving images with scores higher than the threshold to a first DB and other images to a third DB based on the result of comparing the scores of the plurality of images with a threshold, respectively; a step of detecting that the number of images with scores higher than the threshold accumulated in the first DB exceeds a preset number; a step of detecting that, after the exceedance state is detected, the score of a first image not moved to the first DB is higher than the threshold; a step of moving a second image among the images moved to the first DB to a second DB and moving the first image to the first DB; and a step of generating a data set with data extracted from the first DB and the second DB, respectively, when the movement of the plurality of images to any one of the first DB to the third DB is completed.

Inventors

조훈경
이재윤

Assignees

포티투닷 주식회사

Dates

Publication Date: 20260508
Application Date: 20230404

Claims (15)

A step of assigning scores to multiple videos; A step of sequentially moving images with scores higher than the threshold to the first DB and other images to the third DB, based on the result of comparing the scores of the plurality of images with a threshold value; A step of detecting that the number of images with scores higher than the threshold accumulated in the first DB exceeds a preset number; After the above excess state is detected, a step of detecting that the score of the first image not moved to the first DB is higher than the threshold; A step of moving a second image among the images moved to the first DB to the second DB, and moving the first image to the first DB; and A method for generating high-efficiency learning data for autonomous driving, comprising the step of generating a data set using data extracted from the first DB and the second DB, respectively, when the movement of the plurality of images is completed to any one of the first DB to the third DB.
In paragraph 1, The above plurality of images are, A method for generating high-efficiency training data for autonomous driving, wherein a single video is divided into videos of a fixed time length.
In paragraph 1, The above plurality of images are, A method for generating high-efficiency training data for autonomous driving, which is video captured by a camera mounted on a vehicle while the vehicle is in motion.
In paragraph 1, The step of detecting that the number accumulated in the first DB exceeds a preset number is A method for generating high-efficiency training data for autonomous driving, which performs labeling on images stored in the first DB above.
In paragraph 1, The step of moving the above first image to the above first DB is, A method for generating high-efficiency learning data for autonomous driving, wherein an image satisfying exclusion conditions among the images stored in the first DB is determined as a second image and moved to the second DB.
In paragraph 5, Images satisfying the above exclusion conditions are, A method for generating high-efficiency learning data for autonomous driving, wherein the image that records the highest evaluation value when a quantitative evaluation is performed through a preset model among the images stored in the first DB above.
In paragraph 5, Images satisfying the above exclusion conditions are, A method for generating high-efficiency learning data for autonomous driving, wherein images stored in the first DB are embedded to generate a plurality of feature vectors, and a single image is determined based on the result of comparing the feature vectors with each other.
In paragraph 5, Images satisfying the above exclusion conditions are, A method for generating high-efficiency learning data for autonomous driving, wherein images stored in the first DB are classified by a predetermined classification algorithm to generate at least one cluster, and the image is determined based on the distance from the center of the cluster.
In paragraph 1, A method for generating high-efficiency training data for autonomous driving, wherein the maximum capacities of the first DB and the second DB are different.
In paragraph 1, A method for generating high-efficiency training data for autonomous driving, wherein the maximum capacity of the first DB is determined by the computational speed and resources of the device's hardware.
In paragraph 1, The scores assigned to each of the above multiple images are, A method for generating high-efficiency training data for autonomous driving, determined by at least one of the recognition rate and recognition difficulty of an object included in each of the plurality of images above.
In paragraph 1, A method for generating high-efficiency training data for autonomous driving, wherein the number of images that can be stored in the first DB is pre-set.
In paragraph 1, A method for generating high-efficiency training data for autonomous driving, wherein the above dataset consists of all images stored in the first DB and some images stored in the second DB.
A computer-readable recording medium storing a program for executing the method according to paragraph 1 through a computer.
Memory in which at least one program is stored; and By executing at least one of the above programs, the processor performs operations, and The above processor is, Assign scores to multiple videos, and Based on the result of comparing the scores of the aforementioned multiple images with a threshold value, images with scores higher than the threshold value are sequentially moved to the first DB, and the other images are moved to the third DB. Detecting that the number of images with scores higher than the above threshold accumulated in the first DB has exceeded a preset number, After the above excess state is detected, it is detected that the score of the first image that has not been moved to the first DB is higher than the threshold, and Among the images moved to the first DB, the second image is moved to the second DB, and the first image is moved to the first DB. A high-efficiency learning data generation device for autonomous driving that generates a data set using data extracted from the first DB and the second DB, respectively, when the movement of the plurality of images is completed to any one of the first DB to the third DB.

Description

Method for generating high efficiency training data on self-driving and apparatus thereof The present invention relates to a method for generating training data images, and more specifically, to a method for generating high-efficiency training data for autonomous driving and an apparatus for implementing the method. The smartification of vehicles is rapidly progressing due to the convergence of information and communication technology (ICT) and the automotive industry. As a result of this smartification, vehicles are evolving from simple mechanical devices into smart cars, and self-driving is receiving particular attention as a core technology for smart cars. Self-driving is a technology in which an autonomous driving module installed in the vehicle actively controls the driving state, allowing the vehicle to find its way to a destination on its own without the driver having to operate the steering wheel, accelerator pedal, or brakes. To ensure safe autonomous driving, various studies are being conducted on methods for vehicles to accurately recognize pedestrians or other vehicles and calculate the distance to recognized objects during the driving process. However, since the characteristics of objects that may appear on the road while driving are virtually infinite and there are limitations to the processing capabilities of modules installed in autonomous vehicles, no method is currently known to perfectly recognize objects on the road. In the case of object recognition and distance estimation using cameras, a significant amount of distance information is lost because objects from the real 3D world are projected onto a 2D image. In particular, the error is large due to the large deviations in features frequently used for pedestrian position calculation (such as the pedestrian's height or points touching the ground). In the case of object recognition and distance estimation using radar, the ability to rapidly identify and classify objects is poor due to the characteristics of the radio waves operated by the radar; consequently, it is difficult to determine whether an object is a pedestrian or a vehicle. In particular, recognition results tend to be even worse for pedestrians or two-wheeled vehicles (bicycles or motorcycles) on the road because their signal strength is low. Recently, object recognition and distance estimation technologies using LiDAR have been gaining attention due to their relatively high accuracy; however, high-power lasers pose a risk, so LiDAR must operate based on lower-power lasers. Furthermore, unlike the radio waves used by radar, lasers are significantly affected by the surrounding environment, and the excessively high cost of LiDAR sensors is cited as a limitation. The aforementioned background technology is technical information that the inventor possessed for the derivation of the present invention or acquired during the process of deriving the present invention, and it cannot be considered as prior art disclosed to the general public prior to the filing of the present invention. FIGS. 1 to 3 are drawings for explaining an autonomous driving method according to one embodiment. FIGS. 4a and FIGS. 4b are drawings related to a camera that photographs the exterior of a vehicle according to one embodiment. FIG. 5 is a schematic diagram illustrating an object recognition method according to one embodiment. FIG. 6 is a diagram illustrating a labeling process according to an embodiment of the present invention. FIG. 7 is a diagram for conceptually explaining an embodiment of a method for generating learning data according to the present invention. Figure 8 is a diagram illustrating the process of moving a segmented image to the first DB. Figure 9 is a diagram illustrating the process of moving a segmented image to the second DB. Figure 10 is a diagram illustrating the process of moving a segmented image to a third DB. Figure 11 is a diagram that intuitively illustrates the results of the processes of Figures 8 to 10. FIG. 12 is a flowchart illustrating an example of a cuboid acquisition method according to the present invention. FIG. 13 is a block diagram of a learning data generation device according to one embodiment. The present invention is capable of various modifications and may have various embodiments; specific embodiments are illustrated in the drawings and described in detail in the detailed description. The effects and features of the present invention, and the methods for achieving them, will become clear by referring to the embodiments described below in detail together with the drawings. However, the present invention is not limited to the embodiments disclosed below but can be implemented in various forms. Hereinafter, embodiments of the present invention will be described in detail with reference to the attached drawings. When describing with reference to the drawings, identical or corresponding components are given the same reference numerals, and redundant descriptions t