Abstract: The present application discloses a method for acquiring three-dimensional perception information based on external parameters of a roadside camera, and a roadside device. The specific implementation solution is as follows: acquiring a first matching point pair between an image captured by the first camera and an image captured by the second camera, and generating a first rotation matrix, where the first rotation matrix represents a rotation matrix of the first camera in a second camera coordinate system; generating a third rotation matrix according to the first rotation matrix and the second rotation matrix, where the second rotation matrix represents a rotation matrix of the second camera in a world coordinate system, and the third rotation matrix represents a rotation matrix of the first camera in the world coordinate system; generating three-dimensional perception information of the image captured by the first camera according to the third rotation matrix.

Abstract: A method and an apparatus for stabilizing an image, a roadside device, and a cloud control platform are provided. The method may include: calculating, for a current image in an image sequence, a first offset between each image stabilization box of the current image and a corresponding image stabilization box of a template image; adding a first offset meeting a first preset condition to an observation list of a corresponding image stabilization box of the template image until there is an observation list meeting a second preset condition; and determining a position of the signal light of the current image, based on a first offset in an observation list of at least one image stabilization box and a position of the signal light of the template image.

Abstract: The present application discloses a method and an apparatus of obstacle three-dimensional position acquisition for a roadside computing device, and relates to the fields of intelligent transportation, cooperative vehicle infrastructure, and autonomous driving. The method may include: acquiring pixel coordinates of an obstacle in a to-be-processed image; determining coordinates of a bottom surface center point of the obstacle according to the pixel coordinates; acquiring a homography relationship between a ground surface corresponding to the to-be-processed image and a ground surface corresponding to a template image; transforming the coordinates of the bottom surface center point of the obstacle into coordinates on the template image according to the homography relationship; and determining three-dimensional coordinates of the bottom surface center point of the obstacle according to the coordinates obtained by transformation and a ground equation corresponding to the template image.

Abstract: The present disclosure discloses a method and apparatus for detecting an obstacle, and relates to the technical field of intelligent transportation. A specific implementation plan is: acquiring a current image acquired by a camera; inputting the current image into a pre-trained detection model to obtain a position of a detection frame of an obstacle and determine a first pixel coordinate of a grounding point in the current image; determining an offset between the current image and a template image; converting the first pixel coordinate into a world coordinate of the grounding point based on the offset; and outputting the world coordinate of the grounding point as a position of the obstacle in a world coordinate system. This embodiment solves the problem of camera jitter from an image perspective, greatly improves the robustness of the roadside perception system, and saves computing resources.

Abstract: A method and apparatus for jointly calibrating external parameters of multiple cameras. The method includes: determining an overlapped area of angles of view of cameras, and obtaining a 2D verification point set in an image of the overlapped area, the 2D verification point set including at least one verification point; performing external parameter calibration for each camera separately to obtain a current external parameter of the camera; calculating coordinates of a 3D point corresponding to each verification point in a world coordinate system using the current external parameter of each camera separately, and calculating a loss function based on the coordinates of the 3D point, the loss function being used to measure an overlay error of calibration of the cameras under the current external parameter; and performing joint calibration based on the loss function to obtain a target external parameter of each camera.

Abstract: A method for detecting an obstacle, an electronic device, a roadside device and a cloud control platform are provided and relates to the fields of automatic drive and intelligent traffic. The method includes acquiring a current image, wherein the current image includes an obstacle object representing an obstacle located on a predetermined plane; determining a transformation parameter between the current image and a template image based on current coordinates of a plurality of reference points on the predetermined plane in the current image and template coordinates of corresponding reference points for the plurality of reference points in the template image; and determining a position of the obstacle in a world coordinate system utilizing the transformation parameter and pixel coordinates of the obstacle object in the current image.

Abstract: The present application discloses a method and an apparatus of obstacle three-dimensional position acquisition for a roadside computing device, and relates to the fields of intelligent transportation, cooperative vehicle infrastructure, and autonomous driving. The method may include: acquiring pixel coordinates of an obstacle in a to-be-processed image; determining coordinates of a bottom surface center point of the obstacle according to the pixel coordinates; acquiring a homography relationship between a ground surface corresponding to the to-be-processed image and a ground surface corresponding to a template image; transforming the coordinates of the bottom surface center point of the obstacle into coordinates on the template image according to the homography relationship; and determining three-dimensional coordinates of the bottom surface center point of the obstacle according to the coordinates obtained by transformation and a ground equation corresponding to the template image.

Abstract: A method and an apparatus for stabilizing an image, a roadside device, and a cloud control platform are provided. The method may include: calculating, for a current image in an image sequence, a first offset between each image stabilization box of the current image and a corresponding image stabilization box of a template image; adding a first offset meeting a first preset condition to an observation list of a corresponding image stabilization box of the template image until there is an observation list meeting a second preset condition; and determining a position of the signal light of the current image, based on a first offset in an observation list of at least one image stabilization box and a position of the signal light of the template image.

Abstract: The present application discloses a method for acquiring three-dimensional perception information based on external parameters of a roadside camera, and a roadside device. The specific implementation solution is as follows: acquiring a first matching point pair between an image captured by the first camera and an image captured by the second camera, and generating a first rotation matrix, where the first rotation matrix represents a rotation matrix of the first camera in a second camera coordinate system; generating a third rotation matrix according to the first rotation matrix and the second rotation matrix, where the second rotation matrix represents a rotation matrix of the second camera in a world coordinate system, and the third rotation matrix represents a rotation matrix of the first camera in the world coordinate system; generating three-dimensional perception information of the image captured by the first camera according to the third rotation matrix.

Abstract: The present disclosure discloses a method and apparatus for detecting an obstacle, and relates to the technical field of intelligent transportation. A specific implementation plan is: acquiring a current image acquired by a camera; inputting the current image into a pre-trained detection model to obtain a position of a detection frame of an obstacle and determine a first pixel coordinate of a grounding point in the current image; determining an offset between the current image and a template image; converting the first pixel coordinate into a world coordinate of the grounding point based on the offset; and outputting the world coordinate of the grounding point as a position of the obstacle in a world coordinate system. This embodiment solves the problem of camera jitter from an image perspective, greatly improves the robustness of the roadside perception system, and saves computing resources.

Abstract: A method and apparatus for jointly calibrating external parameters of multiple cameras. The method includes: determining an overlapped area of angles of view of cameras, and obtaining a 2D verification point set in an image of the overlapped area, the 2D verification point set including at least one verification point; performing external parameter calibration for each camera separately to obtain a current external parameter of the camera; calculating coordinates of a 3D point corresponding to each verification point in a world coordinate system using the current external parameter of each camera separately, and calculating a loss function based on the coordinates of the 3D point, the loss function being used to measure an overlay error of calibration of the cameras under the current external parameter; and performing joint calibration based on the loss function to obtain a target external parameter of each camera.