KR-20260062249-A - Apparatus and Method For Processing Data Of Artificial Intelligence Learning For Automorus Driving

KR20260062249AKR 20260062249 AKR20260062249 AKR 20260062249AKR-20260062249-A

Abstract

The present invention relates to a data processing device and method for artificial intelligence learning for autonomous driving, comprising: a communication unit for acquiring sensor data from a plurality of sensors equipped in a vehicle; an input unit for receiving a manual annotation for detecting an object included in the sensor data; and a processor that, when the manual annotation is input to the nth frame among a plurality of frames included in the sensor data, automatically detects the object included in the plurality of frames based on the distance between frames of the sensor data and the manual annotation, performs automatic annotation, and stores the data for the automatic annotation as a data set for object recognition. This allows for more effective and rapid processing of data acquired by semi-automating annotation based on multi-sensor fusion SLAM, maintains consistency while minimizing operator subjectivity, and improves accuracy in object recognition.

Inventors

이영원
송광열
노형주

Assignees

한국자동차연구원

Dates

Publication Date: 20260507
Application Date: 20241028

Claims (16)

A communication unit that acquires sensor data from a plurality of sensors equipped in a vehicle; An input unit for receiving a manual annotation for detecting an object included in the sensor data; and A data processing device for artificial intelligence learning for autonomous driving, comprising: a processor that, when the manual annotation is input into the n-th frame (n is a natural number) among the plurality of frames included in the sensor data, automatically detects the object included in the plurality of frames based on the distance between frames of the sensor data and the manual annotation, performs automatic annotation, and stores the data for the automatic annotation as a dataset for object recognition.
In Article 1, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor receives the manual annotation through the input unit for the nth frame containing the object among a plurality of frames.
In Article 1, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor selects m frames containing the object among a plurality of consecutive frames based on the n frames, and performs the automatic annotation on the m frames.
In Article 1, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor estimates the position of the object for the n+1 frame, which is the next frame of the n-th frame, based on the distance traveled between frames, and detects the object included in the n+1 frame based on the manual annotation to perform the automatic annotation.
In Article 1, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor estimates a change in the shape or size of the manual annotation according to at least one of the distance traveled between frames, the installation position and angle of the plurality of sensors, and performs the automatic annotation for the object.
In Article 1, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor fuses the sensor data to generate a map of the surrounding environment and estimates the position of the vehicle, and calculates the distance traveled between frames of the sensor data by calculating the accumulated distance traveled.
In Article 6, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor includes sensor data for LiDAR, a camera, an Inertial Measurement Unit (IMU), and a position sensor (GPS).
In Article 7, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor sets the initial position and direction of the vehicle using sensor data from the position sensor (GPS) and the inertial measurement unit (IMU), extracts feature points of the surrounding environment based on sensor data from the camera and the LiDAR, and mutually matches the feature points to generate the map and estimate the position of the vehicle.
In Article 6, A data processing device for artificial intelligence learning for autonomous driving, characterized in that the processor fuses the sensor data to generate the map based on SLAM (Simultaneous Localization And Mapping).
A step in which a communication unit acquires sensor data from a plurality of sensors equipped in a vehicle; A step in which a processor receives a manual annotation for detecting an object included in the sensor data through an input unit; The above processor, when the manual annotation is input into the n-th frame (n is a natural number) among the plurality of frames included in the sensor data, automatically detects the object included in the plurality of frames based on the inter-frame movement distance of the sensor data and the manual annotation and performs automatic annotation; and A method for processing data for artificial intelligence learning for autonomous driving, comprising the step of the processor storing data for the automatic annotation as a dataset for object recognition.
In Article 10, In the step of receiving the above manual annotation, A method for processing data for artificial intelligence learning for autonomous driving, characterized in that the processor receives the manual annotation through the input unit for the nth frame containing the object among a plurality of frames.
In Article 10, The step of performing the above automatic annotation is, The step of the processor selecting m frames containing the object among a plurality of consecutive frames based on the n frames; and A method for processing data for artificial intelligence learning for autonomous driving, comprising the step of the processor performing the automatic annotation on the m frames.
In Article 10, The step of performing the above automatic annotation is, A step in which the processor estimates the position of the object for the n+1th frame, which is the next frame of the nth frame, based on the distance traveled between frames; The step of the processor detecting the object included in the n+1th frame based on the manual annotation; and A method for processing data for artificial intelligence learning for autonomous driving, characterized by including the step of the processor performing the automatic annotation for the object included in the n+1 frame.
In Article 10, In the step of performing the above automatic annotation, A method for processing data for artificial intelligence learning for autonomous driving, characterized in that the processor estimates a change in the shape or size of the manual annotation according to at least one of the distance traveled between frames, the installation position and angle of the plurality of sensors, and performs the automatic annotation for the object.
In Article 10, Before the step of receiving the above manual annotation, A step in which the above processor fuses the sensor data; The above processor generates a map of the surrounding environment based on the sensor data and estimates the position of the vehicle; and A method for processing data for artificial intelligence learning for autonomous driving, comprising the step of the processor calculating the accumulated distance traveled to calculate the distance traveled between frames of the sensor data.
In Article 10, In the step of acquiring the above sensor data, A method for processing data for artificial intelligence learning for autonomous driving, characterized in that the communication unit acquires sensor data for LiDAR, a camera, an Inertial Measurement Unit (IMU), and a position sensor (GPS).

Description

Apparatus and Method For Processing Data Of Artificial Intelligence Learning For Autonomous Driving The present invention relates to an artificial intelligence learning data processing device and a method for obtaining learning data for autonomous driving of a vehicle. Recently, research on autonomous driving has been actively underway. Autonomous driving refers to a vehicle driving from a starting point to a destination while automatically controlling itself, utilizing road map information, GPS location data, and signals acquired from various sensors. For a vehicle to operate autonomously, technology capable of detecting and recognizing objects from the surrounding environment is required. To achieve this, vehicles employ methods to detect objects using sensor information from cameras, radar, lidar, and other sensors. However, for object recognition and to improve accuracy, many types of labeled data are required, and there is a problem in that building such data requires a significant amount of manpower and time. Accordingly, there is a growing trend of research utilizing artificial intelligence or stereo vision technology to automate annotation. However, annotation automation faces challenges because models that have not been pre-trained exhibit low accuracy, require direct human intervention, and necessitate subjective human judgment to maintain high accuracy, data diversity, and consistency. Furthermore, when a large number of personnel handle the task, annotation automation may result in variations among workers, necessitating separate correction work. Accordingly, there is a need for a method to process autonomous driving training data more efficiently. A related background technology is Korean Patent Publication No. 10-2024-0011067, "Method and apparatus for improving object recognition rate of autonomous vehicles." FIG. 1 is a diagram showing the configuration of a data processing device for artificial intelligence learning for autonomous driving according to one embodiment of the present invention. FIG. 2 is a block diagram briefly illustrating the control configuration of a data processing device according to one embodiment of the present invention. FIG. 3 is a flowchart illustrating the data processing flow of a data processing device according to one embodiment of the present invention. FIG. 4 is an exemplary diagram illustrating a data processing example of a data processing device according to one embodiment of the present invention. FIG. 5 is a flowchart illustrating a data processing method of a data processing device according to an embodiment of the present invention. The present invention will be described below with reference to the attached drawings. In this process, the thickness of lines or the size of components depicted in the drawings may be exaggerated for the sake of clarity and convenience of explanation. Furthermore, the terms described below are defined considering their functions in the present invention, and these may vary depending on the intent or convention of the user or operator. Therefore, the definitions of these terms should be based on the content throughout this specification. FIG. 1 is a diagram showing the configuration of a data processing device for artificial intelligence learning for autonomous driving according to one embodiment of the present invention. As illustrated in FIG. 1, the configuration of a data processing device for artificial intelligence learning for autonomous driving (hereinafter, data processing device) (100) according to the present invention is illustrated. The data processing device (100) can collect sensor data from a plurality of sensors (40) provided in the vehicle (10) and generate a data set for autonomous driving. For example, the data processing device (100) can collect sensor data regarding at least one of the sensors (40) of the vehicle (10), including a lidar (41), a camera (42), an inertial measurement unit (IMU) (43), and a position sensor (GPS) (44). The data processing device (100) can collect data from various sensors in addition to the sensors described. The data processing device (100) is provided outside the vehicle (10) and can receive sensor data from the vehicle (10). Additionally, the data processing device (100) may be installed inside the vehicle (10). The data processing device (100) can collect sensor data of a vehicle (10) based on vehicle-to-vehicle communication (V2V) and vehicle-to-object communication (V2X). Additionally, the data processing device (100) can receive sensor data from multiple vehicles (10) via a network or acquire sensor data through a separate server or database. Furthermore, the data processing device (100) can acquire synchronization data of the sensor data. The data processing device (100) can generate a training data set for object recognition by integrating multiple sensor data. The data processing device (100) can synchronize multiple sensor data based on synchronization data and fus