라이다 및 카메라 시퀀스

6 분

kognic-io 2.5.0부터는 씬 타입(LidarsAndCamerasSequence)을 노출하지 않고 단일 모델로부터 모든 씬 생성을 간소화하는 새로운 씬 모델을 권장하고 있습니다.

LidarsAndCamerasSequence는 카메라 이미지와 라이다 포인트 클라우드의 시퀀스로 구성되며, 각 프레임은 1~~12개의 카메라 이미지와 1~~20개의 포인트 클라우드로 이루어집니다. LidarsAndCamerasSequence 객체의 각 필드가 무엇에 대응하는지에 대한 자세한 문서는 씬 개요 섹션을 참고하세요.

from __future__ import absolute_import

from datetime import datetime
from typing import Optional
from uuid import uuid4

import kognic.io.model.scene.lidars_and_cameras_sequence as LCSM
from examples.calibration.calibration import create_sensor_calibration
from kognic.io.client import KognicIOClient
from kognic.io.logger import setup_logging
from kognic.io.model import CreateSceneResponse, Image, PointCloud


def run(client: KognicIOClient, dryrun: bool = True, **kwargs) -> Optional[CreateSceneResponse]:
    print("Creating Lidar and Camera Sequence Scene...")

    lidar_sensor1 = "lidar"
    cam_sensor1 = "RFC01"
    cam_sensor2 = "RFC02"
    metadata = {"location-lat": 27.986065, "location-long": 86.922623, "vehicle_id": "abg"}

    # Create calibration
    # (Please refer to the API documentation about calibration for more details)
    calibration_spec = create_sensor_calibration(
        f"Collection {datetime.now()}",
        [lidar_sensor1],
        [cam_sensor1, cam_sensor2],
    )
    created_calibration = client.calibration.create_calibration(calibration_spec)

    scene = LCSM.LidarsAndCamerasSequence(
        external_id=f"LCS-example-{uuid4()}",
        frames=[
            LCSM.Frame(
                frame_id="1",
                relative_timestamp=0,
                point_clouds=[
                    PointCloud(
                        filename="./examples/resources/point_cloud_RFL01.las",
                        sensor_name=lidar_sensor1,
                    ),
                ],
                images=[
                    Image(
                        filename="./examples/resources/img_RFC01.jpg",
                        sensor_name=cam_sensor1,
                    ),
                    Image(
                        filename="./examples/resources/img_RFC02.jpg",
                        sensor_name=cam_sensor2,
                    ),
                ],
                metadata={"dut_status": "active"},
            ),
            LCSM.Frame(
                frame_id="2",
                relative_timestamp=100,
                point_clouds=[
                    PointCloud(
                        filename="./examples/resources/point_cloud_RFL02.las",
                        sensor_name=lidar_sensor1,
                    ),
                ],
                images=[
                    Image(
                        filename="./examples/resources/img_RFC11.jpg",
                        sensor_name=cam_sensor1,
                    ),
                    Image(
                        filename="./examples/resources/img_RFC12.jpg",
                        sensor_name=cam_sensor2,
                    ),
                ],
                metadata={"dut_status": "active"},
            ),
        ],
        calibration_id=created_calibration.id,
        metadata=metadata,
    )
    # Create scene
    return client.lidars_and_cameras_sequence.create(scene, dryrun=dryrun, **kwargs)


if __name__ == "__main__":
    setup_logging(level="INFO")
    client = KognicIOClient()

    # Project - Available via `client.project.get_projects()`
    project = "<project-identifier>"

    run(client, project)

dryrun을 사용하여 씬 검증하기

메서드 호출 시 dryrun 파라미터를 true로 설정하면 API를 사용해 씬을 검증하지만 실제로 생성하지는 않습니다.

캘리브레이션 재사용

가능하다면 여러 씬에 대해 동일한 캘리브레이션을 재사용할 수 있으며, 그렇게 하는 것이 좋습니다.

자차(Ego Vehicle) 모션 정보 제공하기

자차 모션(즉, 자차의 위치 및 회전) 정보는 LidarsAndCamerasSeq를 생성할 때 제공할 수 있는 선택적 정보입니다. 이 정보를 제공하면 정지된 객체를 어노테이션하는 데 걸리는 시간을 크게 단축할 수 있습니다. 자차 모션 정보는 씬 내의 각 Frame에 EgoVehicleMotion 객체를 전달함으로써 제공됩니다.

from __future__ import absolute_import

from datetime import datetime
from typing import Optional
from uuid import uuid4

import kognic.io.model.scene.lidars_and_cameras_sequence as LCSM
from examples.calibration.calibration import create_sensor_calibration
from kognic.io.client import KognicIOClient
from kognic.io.logger import setup_logging
from kognic.io.model import CreateSceneResponse, EgoVehiclePose, Image, PointCloud, Position, RotationQuaternion


def run(client: KognicIOClient, dryrun: bool = True, **kwargs) -> Optional[CreateSceneResponse]:
    print("Creating Lidar and Camera Sequence Scene...")

    lidar_sensor1 = "lidar"
    cam_sensor1 = "RFC01"
    cam_sensor2 = "RFC02"
    cam_sensor3 = "RFC03"
    metadata = {"location-lat": 27.986065, "location-long": 86.922623, "vehicle_id": "abg"}

    # Create calibration
    calibration_spec = create_sensor_calibration(f"Collection {datetime.now()}", [lidar_sensor1], [cam_sensor1, cam_sensor2, cam_sensor3])
    created_calibration = client.calibration.create_calibration(calibration_spec)

    scene = LCSM.LidarsAndCamerasSequence(
        external_id=f"LCS-full-example-{uuid4()}",
        frames=[
            LCSM.Frame(
                frame_id="1",
                relative_timestamp=0,
                point_clouds=[
                    PointCloud(filename="./examples/resources/point_cloud_RFL01.las", sensor_name=lidar_sensor1),
                ],
                images=[
                    Image(filename="./examples/resources/img_RFC01.jpg", sensor_name=cam_sensor1),
                    Image(filename="./examples/resources/img_RFC02.jpg", sensor_name=cam_sensor2),
                ],
                metadata={"dut_status": "active"},
                ego_vehicle_pose=EgoVehiclePose(
                    position=Position(x=1.0, y=1.0, z=1.0), rotation=RotationQuaternion(w=0.01, x=1.01, y=1.01, z=1.01)
                ),
            ),
            LCSM.Frame(
                frame_id="2",
                relative_timestamp=500,
                point_clouds=[
                    PointCloud(filename="./examples/resources/point_cloud_RFL02.las", sensor_name=lidar_sensor1),
                ],
                images=[
                    Image(filename="./examples/resources/img_RFC11.jpg", sensor_name=cam_sensor1),
                    Image(filename="./examples/resources/img_RFC12.jpg", sensor_name=cam_sensor2),
                ],
                ego_vehicle_pose=EgoVehiclePose(
                    position=Position(x=2.0, y=2.0, z=2.0), rotation=RotationQuaternion(w=0.01, x=2.01, y=2.01, z=2.01)
                ),
            ),
        ],
        calibration_id=created_calibration.id,
        metadata=metadata,
    )
    # Create scene
    return client.lidars_and_cameras_sequence.create(scene, dryrun=dryrun, **kwargs)


if __name__ == "__main__":
    setup_logging(level="INFO")
    client = KognicIOClient()

    # Project - Available via `client.project.get_projects()`
    project = "<project-identifier>"

    run(client, project=project)

좌표계

position과 rotation 모두 자차 포즈에 있어서는 로컬 좌표계를 기준으로 함에 유의하세요.

셔터 타이밍

Frame 내의 Image에는 셔터 타이밍을 제공할 수 있습니다.

셔터 타이밍은 나노초 단위의 유닉스 타임스탬프 값 쌍이며, 셔터 시작 시각과 종료 시각 두 값 모두 반드시 제공되어야 합니다. 각 프레임의 각 이미지마다 지정됩니다.

from __future__ import absolute_import

import os.path
from datetime import datetime
from typing import Optional
from uuid import uuid4

import kognic.io.model.scene.lidars_and_cameras_sequence as LCSM
from examples.calibration.calibration import create_sensor_calibration
from examples.imu_data.create_imu_data import create_dummy_imu_data
from kognic.io.client import KognicIOClient
from kognic.io.logger import setup_logging
from kognic.io.model import CreateSceneResponse, Image, ImageMetadata, PointCloud


def run(client: KognicIOClient, dryrun: bool = True, **kwargs) -> Optional[CreateSceneResponse]:
    print("Creating Lidar and Camera Sequence Scene...")

    lidar_sensor1 = "RFL01"
    lidar_sensor2 = "RFL02"
    cam_sensor1 = "RFC01"
    cam_sensor2 = "RFC02"
    metadata = {"location-lat": 27.986065, "location-long": 86.922623, "vehicleId": "abg"}
    examples_path = os.path.dirname(__file__)

    # Create calibration
    calibration_spec = create_sensor_calibration(f"Collection {datetime.now()}", [lidar_sensor1, lidar_sensor2], [cam_sensor1, cam_sensor2])
    created_calibration = client.calibration.create_calibration(calibration_spec)

    # Generate IMU data
    ONE_MILLISECOND = 1000000  # one millisecond, expressed in nanos
    start_ts = 1648200140000000000
    end_ts = start_ts + 10 * ONE_MILLISECOND
    imu_data = create_dummy_imu_data(start_timestamp=start_ts, end_timestamp=end_ts, samples_per_sec=1000)

    scene = LCSM.LidarsAndCamerasSequence(
        external_id=f"LCS-full-with-imu-and-shutter-example-{uuid4()}",
        frames=[
            LCSM.Frame(
                frame_id="1",
                unix_timestamp=start_ts + ONE_MILLISECOND,
                relative_timestamp=0,
                point_clouds=[
                    PointCloud(filename=examples_path + "/resources/point_cloud_RFL01.csv", sensor_name=lidar_sensor1),
                    PointCloud(filename=examples_path + "/resources/point_cloud_RFL02.csv", sensor_name=lidar_sensor2),
                ],
                images=[
                    Image(
                        filename=examples_path + "/resources/img_RFC01.jpg",
                        sensor_name=cam_sensor1,
                        metadata=ImageMetadata(
                            shutter_time_start_ns=start_ts + 0.5 * ONE_MILLISECOND, shutter_time_end_ns=start_ts + 1.5 * ONE_MILLISECOND
                        ),
                    ),
                    Image(
                        filename=examples_path + "/resources/img_RFC02.jpg",
                        sensor_name=cam_sensor2,
                        metadata=ImageMetadata(
                            shutter_time_start_ns=start_ts + 0.5 * ONE_MILLISECOND, shutter_time_end_ns=start_ts + 1.5 * ONE_MILLISECOND
                        ),
                    ),
                ],
            ),
            LCSM.Frame(
                frame_id="2",
                unix_timestamp=start_ts + 5 * ONE_MILLISECOND,
                relative_timestamp=4,
                point_clouds=[
                    PointCloud(filename=examples_path + "/resources/point_cloud_RFL11.csv", sensor_name=lidar_sensor1),
                    PointCloud(filename=examples_path + "/resources/point_cloud_RFL12.csv", sensor_name=lidar_sensor2),
                ],
                images=[
                    Image(
                        filename=examples_path + "/resources/img_RFC11.jpg",
                        sensor_name=cam_sensor1,
                        metadata=ImageMetadata(
                            shutter_time_start_ns=start_ts + 4.5 * ONE_MILLISECOND, shutter_time_end_ns=start_ts + 5.5 * ONE_MILLISECOND
                        ),
                    ),
                    Image(
                        filename=examples_path + "/resources/img_RFC12.jpg",
                        sensor_name=cam_sensor2,
                        metadata=ImageMetadata(
                            shutter_time_start_ns=start_ts + 4.5 * ONE_MILLISECOND, shutter_time_end_ns=start_ts + 5.5 * ONE_MILLISECOND
                        ),
                    ),
                ],
            ),
        ],
        calibration_id=created_calibration.id,
        metadata=metadata,
        imu_data=imu_data,
    )
    # Create scene
    return client.lidars_and_cameras_sequence.create(scene, dryrun=dryrun, **kwargs)


if __name__ == "__main__":
    setup_logging(level="INFO")
    client = KognicIOClient()

    # Project - Available via `client.project.get_projects()`
    project = "<project-id>"

    run(client, project=project)

역방향 롤링 셔터

Kognic IO 2.17.0부터 사용 가능합니다.

셔터 타이밍이 설정된 경우, 카메라의 롤링 셔터가 아래에서 위로 진행되는 경우를 위해 Images에 True로 설정할 수 있는 reversed_shutter 플래그가 있습니다. 예를 들어 카메라가 뒤집힌 상태로 장착된 경우, 회전된 이미지의 상단 행의 픽셀이 하단 행의 픽셀보다 나중에 촬영된 것입니다. 어노테이션 도구에서 최상의 투영 품질을 얻으려면 이러한 차이를 아는 것이 중요합니다.

ImageMetadata(
   shutter_time_start_ns=1775128684200000000,
   shutter_time_end_ns=1775128685000000000,
   reversed_shutter=True
 )