Nothing Special   »   [go: up one dir, main page]

WO2022183685A1 - 目标检测方法、电子介质和计算机存储介质 - Google Patents

目标检测方法、电子介质和计算机存储介质 Download PDF

Info

Publication number
WO2022183685A1
WO2022183685A1 PCT/CN2021/111973 CN2021111973W WO2022183685A1 WO 2022183685 A1 WO2022183685 A1 WO 2022183685A1 CN 2021111973 W CN2021111973 W CN 2021111973W WO 2022183685 A1 WO2022183685 A1 WO 2022183685A1
Authority
WO
WIPO (PCT)
Prior art keywords
target
point cloud
point
targets
camera
Prior art date
Application number
PCT/CN2021/111973
Other languages
English (en)
French (fr)
Inventor
郑炜栋
Original Assignee
亿咖通（湖北）科技有限公司
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 亿咖通（湖北）科技有限公司 filed Critical 亿咖通（湖北）科技有限公司
Publication of WO2022183685A1 publication Critical patent/WO2022183685A1/zh

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T19/00—Manipulating 3D models or images for computer graphics
    • G06T19/003—Navigation within 3D models or images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/10—Segmentation; Edge detection
    • G06T7/194—Segmentation; Edge detection involving foreground-background segmentation
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/50—Depth or shape recovery
    • G06T7/55—Depth or shape recovery from multiple images
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T7/00—Image analysis
    • G06T7/80—Analysis of captured images to determine intrinsic or extrinsic camera parameters, i.e. camera calibration
    • G06T7/85—Stereo camera calibration
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06T—IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
    • G06T2207/00—Indexing scheme for image analysis or image enhancement
    • G06T2207/10—Image acquisition modality
    • G06T2207/10028—Range image; Depth image; 3D point clouds

Definitions

  • the present invention relates to the technical field of intelligent vehicle environment perception, in particular to a target detection method, an electronic medium and a computer storage medium.
  • the target detection method based on multi-sensor is one of the current research hotspots.
  • the sensory data output by the camera and the lidar are generally fused at the target level to achieve the perception of the environment. This method improves the accuracy of target detection to a certain extent.
  • it has not done a good fusion at the perceptual data level resulting in the loss of a lot of information in the process of perceptual data fusion, which has an impact on the accuracy of target detection.
  • a target detection method comprising:
  • each target included in the image Detecting each target included in the image to obtain each target detection information of each of the targets, and each of the target detection information includes the pixel position of each of the targets;
  • the second three-dimensional information of each target in the vehicle coordinate system is determined according to each of the first three-dimensional information, and each of the target detection information and each of the second three-dimensional information is output.
  • the projecting the first point cloud into the image coordinate system to obtain the second point cloud in the image coordinate system includes:
  • the determining the view cone point cloud corresponding to each target according to the second point cloud located in each target frame includes:
  • Each of the third point clouds corresponding to each of the fourth point clouds is determined as the view cone point cloud corresponding to each of the targets.
  • performing coordinate transformation on each of the view cone point clouds to obtain a target view cone point cloud corresponding to each of the targets includes:
  • a cone point cloud wherein the central axis is a line connecting the center of the target and the origin of the camera coordinate system;
  • each of the first three-dimensional information includes coordinates of a center point, length, width, height, and heading angle of each of the targets, and it is determined according to each of the first three-dimensional information that each of the targets is in the vehicle
  • the second three-dimensional information in the coordinate system includes:
  • For each of the targets determine the coordinates of each corner of the target according to the coordinates of the center point, the length, width, and height, and the heading angle;
  • the target heading angle of the target is calculated according to the coordinates of each of the fourth corner points, and the coordinates of the fourth center point, the length, width and height and the target heading angle are determined as the second heading angle of the target.
  • the first three-dimensional information of each target is traversed, and the second three-dimensional information of each target in the vehicle coordinate system is determined.
  • the extracting the first target point cloud in each of the target frustum point clouds includes:
  • Each of the target point cloud sets is randomly sampled to obtain each of the first target point clouds.
  • performing coordinate transformation on each of the first target point clouds to obtain a second target point cloud corresponding to each of the targets includes:
  • Coordinate transformation is performed on each of the first target point clouds by using the coordinate average value of each of the first target point clouds to obtain the second target point cloud corresponding to each of the targets.
  • the detecting each target included in the image to obtain each target detection information of each target includes:
  • the image is output to a target detection neural network to detect each of the targets contained in the image and obtain each of the target detection information of each of the targets.
  • an electronic device comprising:
  • the computer program when executed by the processor, causes the electronic device to perform a method as described in any of the above embodiments.
  • a computer storage medium wherein the storage medium stores at least one instruction, at least one segment of program, code reading code set or instruction set, the at least one instruction, at least one segment of A program, code set or instruction set is loaded by the processor and executes the method as described in any of the above embodiments.
  • the image and point cloud that are synchronously collected by the camera and the lidar are respectively acquired, and the point cloud within the field of view of the camera is extracted from the point cloud as the first point cloud.
  • project the first point cloud into the image coordinate system to obtain the second point cloud in the image coordinate system, and detect each target contained in the image to obtain each target detection information of each target, wherein each target detection information includes the target detection information of each target. pixel location.
  • the target frame of each target is determined according to the pixel position of each target, and the view cone point cloud corresponding to each target is determined according to the second point cloud located in each target frame.
  • each view cone point cloud After obtaining the view cone point cloud, coordinate transformation is performed on each view cone point cloud to obtain the target view cone point cloud corresponding to each target. After that, extract the first target point cloud in each target view cone point cloud, perform coordinate transformation on each first target point cloud to obtain a second target point cloud corresponding to each target, and perform fitting and regression on each second target point cloud. Obtain the first three-dimensional information of each target. Finally, the second three-dimensional information of each target in the vehicle coordinate system is determined according to each first three-dimensional information, and each target detection information and each second three-dimensional information are output.
  • FIG. 1 is a schematic flowchart of a target detection method according to one or more embodiments of the present invention.
  • FIG. 2 is a schematic diagram of a camera and a lidar synchronously collecting images and point clouds according to one or more embodiments of the present invention
  • FIG. 3 is a schematic diagram of an image according to one or more embodiments of the present invention.
  • FIG. 4 is a schematic diagram of an image and camera coordinate system according to one or more embodiments of the present invention.
  • FIG. 5 is a schematic diagram of a target and camera coordinate system according to one or more embodiments of the present invention.
  • FIG. 6 is a schematic structural block diagram of an electronic device according to one or more embodiments of the present invention.
  • FIG. 1 is a schematic flowchart of a target detection method according to one or more embodiments of the present invention. As shown in Figure 1, the method may include at least the following steps:
  • Step S102 Acquire the image and point cloud synchronously collected by the camera and the lidar respectively, and extract the point cloud located within the field of view of the camera from the point cloud as the first point cloud.
  • Step S104 Project the first point cloud into the image coordinate system to obtain a second point cloud in the image coordinate system.
  • Step S106 Detect each target included in the image to obtain each target detection information of each target, and each target detection information includes a pixel position of each target.
  • Step S108 Determine the target frame of each target according to the pixel position of each target, and determine the view cone point cloud corresponding to each target according to the second point cloud located in each target frame.
  • Step S110 Perform coordinate transformation on each view cone point cloud to obtain a target view cone point cloud corresponding to each target.
  • Step S112 extracting a first target point cloud in each target viewing cone point cloud, and performing coordinate transformation on each first target point cloud to obtain a second target point cloud corresponding to each target.
  • Step S114 Fitting and regressing each second target point cloud to obtain the first three-dimensional information of each target.
  • Step S116 Determine second three-dimensional information of each target in the vehicle coordinate system according to each first three-dimensional information, and output each target detection information and each second three-dimensional information.
  • the image and point cloud that are synchronously collected by the camera and the lidar are respectively acquired, and the point cloud located within the field of view of the camera is extracted from the point cloud as the first point cloud.
  • the first point cloud is projected into the image coordinate system to obtain the second point cloud in the image coordinate system, and each target included in the image is detected to obtain each target detection information of each target.
  • the target frame of each target is determined according to the pixel position of each target, and the view cone point cloud corresponding to each target is determined according to the second point cloud located in each target frame.
  • coordinate transformation is performed on each view cone point cloud to obtain the target view cone point cloud corresponding to each target.
  • each target point cloud in each target view cone point cloud After that, extract the first target point cloud in each target view cone point cloud, perform coordinate transformation on each first target point cloud to obtain a second target point cloud corresponding to each target, and perform fitting and regression on each second target point cloud. Obtain the first three-dimensional information of each target. Finally, the second three-dimensional information of each target in the vehicle coordinate system is determined according to each first three-dimensional information, and each target detection information and each second three-dimensional information are output. According to one or more embodiments of the present invention, by fully fusing the data collected by the camera and the data collected by the lidar, accurate three-dimensional information of objects around the smart vehicle can be obtained, so that during the autonomous driving journey of the smart vehicle, the According to the three-dimensional information, the correct path is planned, which greatly guarantees safe driving.
  • FIG. 2 is a schematic diagram of a camera and a lidar synchronously acquiring images and point clouds according to one or more embodiments of the present invention.
  • the trigger time of the camera and the lidar can be synchronized to the same time axis to ensure that the images and point clouds collected by the camera and the lidar synchronously are obtained respectively.
  • the time axis of the lidar can be used as the benchmark, and the trigger time of the camera can be controlled to be consistent with the trigger time of the lidar as much as possible, for example, the camera and the lidar can be triggered at the same time every 10ms.
  • the camera is calibrated first.
  • the internal parameters are obtained to obtain the internal parameter matrix, and then the camera and the lidar are jointly calibrated with the camera as a reference point to obtain the first coordinate transformation matrix from the lidar to the camera.
  • the first coordinate transformation matrix is used to perform coordinate transformation on the first point cloud to obtain the third point cloud in the camera coordinate system, and then the internal parameter matrix is used to perform coordinate projection on the third point cloud to obtain the second point cloud in the image coordinate system.
  • the intrinsic parameter matrix f x and f y represent the horizontal and vertical focal lengths of the camera respectively, and u and v represent the optical axis projection coordinates in the image coordinate system, that is, the principal point coordinates.
  • the image coordinate system is generally a two-dimensional coordinate system with the upper left corner as the origin, the positive x direction to the right, and the positive y direction to the downward, and the unit is usually "pixel".
  • the first coordinate transformation matrix Among them, Ref '(e, f 1, 2, 3) represents the rotation matrix element of the lidar coordinate system relative to the camera coordinate system.
  • the lidar coordinate system generally takes the laser emission center as the coordinate origin, and the upward direction is the positive direction of the z-axis , the x- and y-axes form a plane.
  • the camera coordinate system generally takes the center of the main optical axis of the lens as the origin, the right direction is the x positive direction, the downward direction is the y positive direction, and the forward direction is the z positive direction.
  • the vector [T 11′ , T 12′ , T 12′ ] b represents the translation relationship between the lidar coordinate system and the camera coordinate system, and b represents the transposition symbol.
  • the image may be output to a target detection neural network to detect each target contained in the image and obtain target detection information of each target , the target detection information includes, for example, classification information, pixel position, and confidence.
  • FIG. 3 is a schematic diagram of an image in accordance with one or more embodiments of the present invention. As shown in Figure 3, the target detection information is as follows:
  • Object i [box(px, py, h, w), class, confidence]
  • Object i is the target
  • i represents the serial number of the target
  • box(px, py, h, w) is the pixel position parameter of the target
  • (px, py) is the pixel coordinate of the target
  • h, w are the height of the target box, respectively and width
  • class is classification information
  • confidence is confidence.
  • the target frame of each target can be determined according to each pixel position, and then the second point cloud located in each target frame is randomly sampled to obtain the fourth point corresponding to each target. cloud, and then each third point cloud corresponding to each fourth point cloud is determined as a view cone point cloud corresponding to each target.
  • FIG. 4 is a schematic diagram of an image and camera coordinate system in accordance with one or more embodiments of the present invention.
  • Figure 4 in order to make the coordinate distribution of all the view cone point clouds similar to the greatest extent for the subsequent point cloud segmentation network processing, after obtaining the view cone point cloud, first determine the center axis of each target and the camera coordinate system. The included angle ⁇ between the Z axes, and then use each included angle ⁇ to perform coordinate transformation on each view cone point cloud to obtain the first view cone point cloud corresponding to each target, where the central axis is the center of the target (cx, cy) and The connection between the origins of the camera coordinate system. Finally, the Y-axis average value of each first view cone point cloud is calculated, and the coordinate transformation of each first view cone point cloud is performed by using the average value of each Y-axis to obtain the target view cone point cloud corresponding to each target.
  • the specific process of converting the view frustum point cloud into the target view frustum point cloud may be as follows:
  • each target view cone point cloud can be output to the point cloud segmentation network to obtain the first probability that each point in each target view cone point cloud belongs to the foreground and the second probability that it belongs to the background.
  • points with a first probability greater than the second probability are selected to form a target point cloud set, and each target point cloud set is randomly sampled to obtain each first target point cloud.
  • each target point cloud set is randomly sampled to obtain each first target point cloud.
  • the average value of the coordinates of the first target point cloud is as follows: Among them, k1 is the number of point clouds of the first target point cloud.
  • the first three-dimensional information of the target can be obtained by outputting the second target point cloud to the point cloud fitting regression network.
  • the form of the first three-dimensional information may be:
  • (x, y, z) are the coordinates of the center point of the target
  • (l, w, h) are the length, width, and height of the target
  • (heading_angle) is the heading angle of the target.
  • FIG. 5 is a schematic diagram of a target and camera coordinate system in accordance with one or more embodiments of the present invention.
  • the external parameters of the camera are calibrated with the inertial measurement device of the vehicle as a reference point to obtain the external parameter matrix.
  • the first three-dimensional information includes the coordinates, length, width, height and heading angle of the center point of the target.
  • the coordinates of each corner point of the target are determined according to the coordinates of the center point, the length, width, and height, and the heading angle, wherein the corner point usually refers to an extreme point, that is, a point that is particularly prominent in some aspects.
  • the corner points may refer to the corner points between lines, as shown in FIG. 5 , for example, if the target is a cuboid, each corner point of the target refers to each vertex of the cuboid. Then, the coordinates of the center point and each corner point are converted by using the coordinate average value of the first target point cloud to obtain the coordinates of the first center point and each first corner point. Then, the coordinates of the first center point and each of the first corner points are converted by using the average value of the Y-axis of the point cloud of the first view cone to obtain the coordinates of the second center point and each of the second corner points.
  • the target heading angle of the target is calculated according to the coordinates of each fourth corner point, and the coordinates, length, width, height and target heading angle of the fourth center point are determined as the second three-dimensional information of the target, and the first three-dimensional information of each target is traversed to determine The second three-dimensional information of each target in the vehicle coordinate system.
  • pos(x, y, z) is the coordinates of the center point of the target in the vehicle coordinate system
  • size(l, w, h) is the length, width, and height of the target
  • heading_angle1 is the heading angle of the target.
  • FIG. 6 is a schematic structural block diagram of an electronic device according to one or more embodiments of the present invention.
  • the electronic device 600 includes a processor 610 and a memory 620 storing a computer program 621.
  • the computer program 621 is executed by the processor 610, the electronic device 600 is caused to perform the method of any of the above-described embodiments.
  • one or more implementations of the present invention further provide a computer storage medium, in which the storage medium stores at least one instruction, at least one program, code reading code set or instruction set, at least one instruction, at least one program , a code set or an instruction set is loaded by the processor and executes the method as described in any of the above embodiments.
  • One or more implementations of the present invention provide a method for object detection based on an in-vehicle camera and a lidar.
  • the image and point cloud that are synchronously collected by the camera and the lidar are respectively acquired, and the point cloud located within the field of view of the camera is extracted from the point cloud as the first point cloud.
  • project the first point cloud into the image coordinate system to obtain the second point cloud in the image coordinate system, and detect each target contained in the image to obtain each target detection information of each target, wherein each target detection information includes the target detection information of each target. pixel location.
  • the target frame of each target is determined according to the pixel position of each target, and the view cone point cloud corresponding to each target is determined according to the second point cloud located in each target frame.
  • coordinate transformation is performed on each view cone point cloud to obtain the target view cone point cloud corresponding to each target.
  • extract the first target point cloud in each target view cone point cloud perform coordinate transformation on each first target point cloud to obtain a second target point cloud corresponding to each target, and perform fitting and regression on each second target point cloud.
  • the second three-dimensional information of each target in the vehicle coordinate system is determined according to each first three-dimensional information, and each target detection information and each second three-dimensional information are output.
  • the present invention by fully fusing the data collected by the camera and the data collected by the lidar, accurate three-dimensional information of objects around the smart vehicle can be obtained, so that during the autonomous driving journey of the smart vehicle, the According to the three-dimensional information, the correct path is planned, which greatly guarantees safe driving.

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Software Systems (AREA)
  • General Engineering & Computer Science (AREA)
  • Computer Hardware Design (AREA)
  • Computer Graphics (AREA)
  • Remote Sensing (AREA)
  • Radar, Positioning & Navigation (AREA)
  • Image Analysis (AREA)
  • Length Measuring Devices By Optical Means (AREA)

Abstract

本公开提供了一种基于车载摄像头和激光雷达的目标检测方法，其包括：分别获取摄像头和激光雷达同步采集到的图像和点云，并且从点云中提取位于摄像头的视场角内的点云作为第一点云；将第一点云投影至图像坐标系中得到图像坐标系下的第二点云；检测图像中包含的各目标得到各目标的各目标检测信息；根据各像素位置确定各目标的目标框，根据位于各目标框内的第二点云确定与各目标对应的视锥点云；对各视锥点云进行坐标转换得到与各目标对应的目标视锥点云；提取各目标视锥点云中的第一目标点云，根据第一目标点云得到各目标的第一三维信息；根据各第一三维信息确定各目标在车辆坐标系下的第二三维信息，输出各目标检测信息和各第二三维信息。

Description

目标检测方法、电子介质和计算机存储介质 技术领域
本发明涉及智能汽车环境感知技术领域，特别是涉及一种目标检测方法、电子介质和计算机存储介质。
背景技术
随着时代的发展、技术的进步，如今，智能汽车成为交通运输工具领域的热门研究话题。对于智能汽车，其要具备可靠的实时环境感知能力和理解周围环境的能力，这样才能规划出正确路径、保证安全驾驶。
现有技术中，很多环境感知方法是基于单传感器的，但是单传感器检测方法无法突破传感器本身的数据特征，导致环境感知能力差，目标检测不准。因此基于多传感器的目标检测方法是目前的研究热点之一。对于基于多传感器的目标检测方法，一般是对摄像头和激光雷达各自输出的感知数据做目标级别的融合以达到对环境的感知，这样的方法在一定程度上提高了目标检测的准确性。但是却并没有在感知数据层面做到很好的融合，导致感知数据融合过程中损失了大量信息，对目标检测的准确性产生了影响。
发明内容
根据本发明的一个方面，提供了一种目标检测方法，其包括：
分别获取摄像头和激光雷达同步采集到的图像和点云，并且从所述点云中提取位于所述摄像头的视场角内的点云作为第一点云；
将所述第一点云投影至图像坐标系中得到所述图像坐标系下的第二点云；
检测所述图像中包含的各目标得到各所述目标的各目标检测信息，各所述目标检测信息包括各所述目标的像素位置；
根据各所述目标的所述像素位置确定各所述目标的目标框，根据位于各所述目标框内的所述第二点云确定与各所述目标对应的视锥点云；
对各所述视锥点云进行坐标转换得到与各所述目标对应的目标视锥点云；
提取各所述目标视锥点云中的第一目标点云，对各所述第一目标点云进行坐标转换得到与各所述目标对应的第二目标点云；
对各所述第二目标点云进行拟合回归得到各所述目标的第一三维信息；
根据各所述第一三维信息确定各所述目标在车辆坐标系下的第二三维信息，输出各所述目标检测信息和各所述第二三维信息。
根据一个或多个实施例，所述将所述第一点云投影至图像坐标系中得到所述图像坐标系下的第二点云，包括：
标定所述摄像头的内参数得到内参数矩阵；
以所述摄像头为参考点联合标定所述摄像头和所述激光雷达得到所述激光雷达到所述摄像头的第一坐标转换矩阵；
利用所述第一坐标转换矩阵对所述第一点云进行坐标转换得到摄像头坐标系下的第三点云；
利用所述内参数矩阵对所述第三点云进行坐标投影得到所述图像坐标系下的所述第二点云。
根据一个或多个实施例，所述根据位于各所述目标框内的所述第二点云确定与各所述目标对应的视锥点云包括：
对位于各所述目标框内的所述第二点云进行随机采样，得到与各所述目标对应的第四点云；
将与各所述第四点云对应的各所述第三点云确定为与各所述目标对应的所述视锥点云。
根据一个或多个实施例，所述对各所述视锥点云进行坐标转换得到与各所述目标对应的目标视锥点云包括：
确定各所述目标的中心轴与所述摄像头坐标系的Z轴间的夹角，利用各所述夹角对各所述视锥点云进行坐标转换得到与各所述目标对应的第一视锥点云，其中，所述中心轴为所述目标的中心与所述摄像头坐标系的原点间的连线；
计算各所述第一视锥点云的Y轴平均值，利用各所述Y轴平均值对各所述第一视锥点云进行坐标转换得到与各所述目标对应的所述目标视锥点云。
根据一个或多个实施例，各所述第一三维信息包括各所述目标的中心点的坐标、长宽高及航向角，所述根据各所述第一三维信息确定各所述目标在车辆坐标系下的第二三维信息包括：
以车辆的惯性测量装置为参考点标定所述摄像头的外参数得到外参数矩阵；
对于每一所述目标，根据所述中心点的坐标、所述长宽高及所述航向角确定所述目标的各角点的坐标；
利用所述第一目标点云的坐标平均值对所述中心点和各所述角点的坐标进 行转换，得到第一中心点和各第一角点的坐标；
利用所述第一视锥点云的Y轴平均值对所述第一中心点和各所述第一角点的坐标进行转换，得到第二中心点和各第二角点的坐标；
利用所述目标的中心轴与所述摄像头坐标系的Z轴间的所述夹角对所述第二中心点和各所述第二角点的坐标进行转换，得到第三中心点和各第三角点的坐标；
利用所述外参数矩阵对所述第三中心点和各所述第三角点的坐标进行转换，得到第四中心点和各第四角点的坐标；
根据各所述第四角点的坐标计算所述目标的目标航向角，将所述第四中心点的坐标、所述长宽高及所述目标航向角确定为所述目标的所述第二三维信息；
遍历各所述目标的所述第一三维信息，确定各所述目标在所述车辆坐标系下的所述第二三维信息。
根据一个或多个实施例，所述提取各所述目标视锥点云中的第一目标点云包括：
将各所述目标视锥点云分别输出至点云分割网络得到各所述目标视锥点云中各点属于前景的第一概率和属于背景的第二概率；
对于各所述目标视锥点云，选取所述第一概率大于所述第二概率的所述点组成目标点云集合；
对各所述目标点云集合进行随机采样得到各所述第一目标点云。
根据一个或多个实施例，所述对各所述第一目标点云进行坐标转换得到与各所述目标对应的第二目标点云包括：
计算各所述第一目标点云的坐标平均值；
利用各所述第一目标点云的坐标平均值对各所述第一目标点云进行坐标转换，得到与各所述目标对应的所述第二目标点云。
根据一个或多个实施例，所述检测所述图像中包含的各目标得到各所述目标的各目标检测信息包括：
将所述图像输出至目标检测神经网络中以检测所述图像包含的各所述目标并得到各所述目标的各所述目标检测信息。
根据本发明的另一个方面，还提供了一种电子设备，其包括：
处理器；
存储有计算机程序的存储器；
当所述计算机程序被所述处理器运行时，导致所述电子设备执行如上述任 意实施例所述的方法。
根据本发明的又一个方面，还提供了一种计算机存储介质，其中，所述存储介质中存储有至少一条指令、至少一段程序、代读码集或指令集，所述至少一条指令、至少一段程序、代码集或指令集由处理器加载并执行如上述任意实施例所述的方法。
在本发明的一个或多个实施例中，首先分别获取摄像头和激光雷达同步采集到的图像和点云并从点云中提取出位于摄像头视场角内的点云作为第一点云。之后将第一点云投影至图像坐标系中得到图像坐标系下的第二点云，并且检测图像中包含的各目标得到各目标的各目标检测信息，其中，各目标检测信息包括各目标的像素位置。之后根据各目标的像素位置确定各目标的目标框，并且根据位于各目标框内的第二点云确定与各目标对应的视锥点云。在得到视锥点云后，再对各视锥点云进行坐标转换得到与各目标对应的目标视锥点云。之后再提取各目标视锥点云中的第一目标点云，对各第一目标点云进行坐标转换得到与各目标对应的第二目标点云，对各第二目标点云进行拟合回归得到各目标的第一三维信息。最后根据各第一三维信息确定各目标在车辆坐标系下的第二三维信息，输出各目标检测信息和各第二三维信息。
根据一个或多个实施例，通过将摄像头采集数据和激光雷达采集数据进行充分融合，从而可以得到智能车辆周围物体的精准的三维信息，从而在智能车辆自动驾驶行程中，可以根据周围物体的三维信息而规划出正确的路径，极大地保障了安全驾驶。
根据下文结合附图对本发明具体实施例的详细描述，本领域技术人员将会更加明了上述以及其他优点和特征。
附图说明
后文将参照附图以示例性而非限制性的方式详细描述本发明的一些具体实施例。附图中相同的附图标记标示了相同或类似的部件或部分。本领域技术人员应该理解，这些附图未必是按比例绘制的。附图中：
图1是根据本发明一个或多个实施例的目标检测方法的流程示意图；
图2是根据本发明一个或多个实施例的摄像头和激光雷达同步采集图像和点云的示意图；
图3是根据本发明一个或多个实施例的图像的示意图；
图4是根据本发明一个或多个实施例的图像和摄像头坐标系的示意图；
图5是根据本发明一个或多个实施例的目标和摄像头坐标系的示意图；
图6是根据本发明一个或多个实施例的电子设备的示意性结构框图。
具体实施方式
下面将参照附图更详细地描述本公开的示例性实施例。虽然附图中显示了本公开的示例性实施例，然而应当理解，可以以各种形式实现本公开而不应被这里阐述的实施例所限制。相反，提供这些实施例是为了能够更透彻地理解本公开，并且能够将本公开的范围完整的传达给本领域的技术人员。
需要说明的是，在不冲突的前提下本发明实施例及可选实施例中的技术特征可以相互结合。
图1是根据本发明一个或多个实施例的目标检测方法的流程示意图。如图1所示，该方法可以至少包括如下步骤：
步骤S102：分别获取摄像头和激光雷达同步采集到的图像和点云，并且从点云中提取位于摄像头的视场角内的点云作为第一点云。
步骤S104：将第一点云投影至图像坐标系中得到图像坐标系下的第二点云。
步骤S106：检测图像中包含的各目标得到各目标的各目标检测信息，各目标检测信息包括各目标的像素位置。
步骤S108：根据各目标的像素位置确定各目标的目标框，根据位于各目标框内的第二点云确定与各目标对应的视锥点云。
步骤S110：对各视锥点云进行坐标转换得到与各目标对应的目标视锥点云。
步骤S112：提取各目标视锥点云中的第一目标点云，对各第一目标点云进行坐标转换得到与各目标对应的第二目标点云。
步骤S114：对各第二目标点云进行拟合回归得到各目标的第一三维信息。
步骤S116：根据各第一三维信息确定各目标在车辆坐标系下的第二三维信息，输出各目标检测信息和各第二三维信息。
在本发明一个或多个实施例中，首先分别获取摄像头和激光雷达同步采集到的图像和点云并从点云中提取出位于摄像头视场角内的点云作为第一点云。之后将第一点云投影至图像坐标系中得到图像坐标系下的第二点云，并且检测图像中包含的各目标得到各目标的各目标检测信息。之后根据各目标的像素位置确定各目标的目标框，并且根据位于各目标框内的第二点云确定与各目标对应的视锥点云。在得到视锥点云后，再对各视锥点云进行坐标转换得到与各目标对应的目标视锥点云。之后再提取各目标视锥点云中的第一目标点云，对各 第一目标点云进行坐标转换得到与各目标对应的第二目标点云，对各第二目标点云进行拟合回归得到各目标的第一三维信息。最后根据各第一三维信息确定各目标在车辆坐标系下的第二三维信息，输出各目标检测信息和各第二三维信息。根据本发明一个或多个实施例，通过将摄像头采集数据和激光雷达采集数据进行充分融合，从而可以得到智能车辆周围物体的精准的三维信息，从而在智能车辆自动驾驶行程中，可以根据周围物体的三维信息而规划出正确的路径，极大地保障了安全驾驶。
图2是根据本发明一个或多个实施例的摄像头和激光雷达同步采集图像和点云的示意图。如图2所示，可以将摄像头和激光雷达的触发时间同步到同一时间轴以保证分别获取摄像头和激光雷达同步采集到的图像和点云。例如，可以以激光雷达的时间轴为基准，控制摄像头的触发时间尽量与激光雷达的触发时间一致，例如每隔10ms同时触发摄像头和激光雷达。
在得到摄像头和激光雷达同步采集到的图像和点云并且从点云中提取出位于摄像头的视场角内的第一点云后，接下来，在本发明一些实施例中，先标定摄像头的内参数得到内参数矩阵，然后以摄像头为参考点联合标定摄像头和激光雷达得到激光雷达到摄像头的第一坐标转换矩阵。最后利用第一坐标转换矩阵对第一点云进行坐标转换得到摄像头坐标系下的第三点云，再利用内参数矩阵对第三点云进行坐标投影得到图像坐标系下的第二点云。
根据一个实施例，内参数矩阵
Figure PCTCN2021111973-appb-000001
其中，f x、f y分别表示摄像头的横、纵焦距，u、v表示图像坐标系中光轴投影坐标即主点坐标。对于电脑上存储的照片或图像，图像坐标系一般是以左上角为原点，向右为x正方向，向下为y正方向的二维坐标系，单位常用“像素”。
第一坐标转换矩阵
Figure PCTCN2021111973-appb-000002
其中，R ef′(e，f＝1，2，3)表示激光雷达坐标系相对于摄像头坐标系的旋转矩阵元素，激光雷达坐标系一般以激光发射中心为坐标原点，向上为z轴正方向，x轴和y轴构成平面。摄像头坐标系一般以镜头主光轴中心为原点，向右为x正方向，向下为y正方向，向前为z正方向。向量[T 11′，T 12′，T 12′] b表示激光雷达坐标系相对于摄像头坐标系的平移关系，b表示转置符号。
利用上述矩阵将第一点云转换为第二点云的具体过程如下：
Figure PCTCN2021111973-appb-000003
Figure PCTCN2021111973-appb-000004
其中，
Figure PCTCN2021111973-appb-000005
表示第一点云的坐标，
Figure PCTCN2021111973-appb-000006
表示第三点云的坐标，
Figure PCTCN2021111973-appb-000007
表示第二点云的坐标。
在将第一点云转换为第二点云后，接下来，在本发明一些实施例中，可以将图像输出至目标检测神经网络中以检测图像包含的各目标并得到各目标的目标检测信息，目标检测信息例如包括分类信息、像素位置及置信度等。
图3是根据本发明一个或多个实施例的图像的示意图。如图3所示，目标检测信息如下：
Object i＝[box(px，py，h，w)，class，confidence]
其中，Object i为目标，i代表目标的序号，box(px，py，h，w)为目标的像素位置参数，(px，py)为目标的像素坐标，h、w分别为目标框的高和宽，class为分类信息，confidence为置信度。
在得到各目标的目标检测信息后，接下来，可以根据各像素位置确定各目标的目标框，然后对位于各目标框内的第二点云进行随机采样，得到与各目标对应的第四点云，然后将与各第四点云对应的各第三点云确定为与各目标对应的视锥点云。
图4是根据本发明一个或多个实施例的图像和摄像头坐标系的示意图。如图4所示，为了使所有视锥点云的坐标分布最大程度上相似以便于后续点云分割网络处理，因此在得到视锥点云后，首先确定各目标的中心轴与摄像头坐标系的Z轴间的夹角θ，然后利用各夹角θ对各视锥点云进行坐标转换得到与各目标对应的第一视锥点云，其中，中心轴为目标的中心(cx，cy)与摄像头坐标系的原点间的连线。最后计算各第一视锥点云的Y轴平均值，利用各Y轴的平均值对各第一视锥点云进行坐标转换得到与各目标对应的目标视锥点云。
根据一个实施例，将视锥点云转换为目标视锥点云的具体过程可以如下：
Figure PCTCN2021111973-appb-000008
Figure PCTCN2021111973-appb-000009
其中，
Figure PCTCN2021111973-appb-000010
表示第一视锥点云的坐标，
Figure PCTCN2021111973-appb-000011
表示目标视锥点云的坐标。
得到目标视锥点云后，接下来，可以将各目标视锥点云分别输出至点云分割网络得到各目标视锥点云中各点属于前景的第一概率和属于背景的第二概率。对于各目标视锥点云，然后选取第一概率大于第二概率的点组成目标点云集合，对各目标点云集合进行随机采样得到各第一目标点云。
由于后续提到的点云拟合回归网络对输入的点云数量有要求，因此，此处对各目标点云集合进行随机采样得到各第一目标点云。
为了使第一目标点云的坐标分布最大程度上相似以便于后续点云拟合回归网络处理。接下来，计算各第一目标点云的坐标平均值，然后利用各第一目标点云的坐标平均值对各第一目标点云进行坐标转换得到与各目标对应的第二目标点云。
根据一个实施例，第一目标点云的坐标的平均值如下：
Figure PCTCN2021111973-appb-000012
Figure PCTCN2021111973-appb-000013
其中，k1为第一目标点云的点云个数。
将第一目标点云转换为第二目标点云的具体过程如下：
Figure PCTCN2021111973-appb-000014
其中，
Figure PCTCN2021111973-appb-000015
为第二目标点云的坐标。
将第二目标点云输出至点云拟合回归网络中可以得到目标的第一三维信息。第一三维信息的形式可以为：
[x，y，z，l，w，h，heading_angle]
其中，(x，y，z)为目标的中心点的坐标，(l，w，h)为目标的长宽高，(heading_angle)为目标的航向角。
图5是根据本发明一个或多个实施例的目标和摄像头坐标系的示意图。如图5所示，为了得到目标在车辆坐标系下的第二三维信息，接下来，首先以车 辆的惯性测量装置为参考点标定摄像头的外参数得到外参数矩阵。第一三维信息包括目标的中心点的坐标、长宽高及航向角。对于每一目标，根据中心点的坐标、长宽高及航向角确定目标的各角点的坐标，其中，角点通常指极值点，即某方面特别突出的点，在本发明一些实施例中，角点可以指的是线与线之间的角点，如图5所示，例如目标为长方体状，则目标的各角点指的是长方体的各顶点。之后利用第一目标点云的坐标平均值对中心点和各角点的坐标进行转换，得到第一中心点和各第一角点的坐标。再利用第一视锥点云的Y轴的平均值对第一中心点和各第一角点的坐标进行转换，得到第二中心点和各第二角点的坐标。再利用目标的中心轴与摄像头坐标系的Z轴间的夹角对第二中心点和各第二角点的坐标进行转换，得到第三中心点和各第三角点的坐标。再利用外参数矩阵对第三中心点和各第三角点的坐标进行转换，得到第四中心点和各第四角点的坐标。最后根据各第四角点的坐标计算目标的目标航向角，将第四中心点的坐标、长宽高及目标航向角确定为目标的第二三维信息，遍历各目标的第一三维信息，确定各目标在车辆坐标系下的第二三维信息。
根据一个实施例，外参数矩阵
Figure PCTCN2021111973-appb-000016
其中，R ij(i，j＝1，2，3)表示旋转矩阵，[T 11，T 12，T 13] b表示平移向量。
将目标的中心点和角点的坐标分别转换为车辆坐标系下的第四中心点和第四角点的坐标的具体过程如下：
Figure PCTCN2021111973-appb-000017
Figure PCTCN2021111973-appb-000018
Figure PCTCN2021111973-appb-000019
Figure PCTCN2021111973-appb-000020
最终输出的信息的形式如下：
Figure PCTCN2021111973-appb-000021
其中，pos(x，y，z)为目标的中心点在车辆坐标系下的坐标，size(l，w，h)为目标的长、宽、高，heading_angle1为目标航向角。
基于同一发明构思，本发明的一个或多个实施例还提出了一种电子设备600。图6是根据本发明一个或多个实施例的电子设备的示意性结构框图。如图6所示，电子设备600包括：处理器610和存储有计算机程序621的存储器620.当计算机程序621被处理器610运行时，导致电子设备600执行如上述任意实施例的方法。
基于同一发明构思，本发明的一个或多个实施还提出了一种计算机存储介质，存储介质中存储有至少一条指令、至少一段程序、代读码集或指令集，至少一条指令、至少一段程序、代码集或指令集由处理器加载并执行如上述任意实施例所述的方法。
本发明的一个或多个实施提供了一种基于车载摄像头和激光雷达的目标检测方法。在本发明一个或多个实施例中，首先分别获取摄像头和激光雷达同步采集到的图像和点云并从点云中提取出位于摄像头视场角内的点云作为第一点云。之后将第一点云投影至图像坐标系中得到图像坐标系下的第二点云，并且检测图像中包含的各目标得到各目标的各目标检测信息，其中，各目标检测信息包括各目标的像素位置。之后根据各目标的像素位置确定各目标的目标框，并且根据位于各目标框内的第二点云确定与各目标对应的视锥点云。在得到视锥点云后，再对各视锥点云进行坐标转换得到与各目标对应的目标视锥点云。之后再提取各目标视锥点云中的第一目标点云，对各第一目标点云进行坐标转换得到与各目标对应的第二目标点云，对各第二目标点云进行拟合回归得到各目标的第一三维信息。最后根据各第一三维信息确定各目标在车辆坐标系下的第二三维信息，输出各目标检测信息和各第二三维信息。根据本发明一个或多个实施例，通过将摄像头采集数据和激光雷达采集数据进行充分融合，从而可以得到智能车辆周围物体的精准的三维信息，从而在智能车辆自动驾驶行程中，可以根据周围物体的三维信息而规划出正确的路径，极大地保障了安全驾驶。
尽管已经针对有限数量的实施例描述了本发明，但是受益于本公开的本领域技术人员将理解，可以设计其他实施例而不脱离本文所公开的本发明的范围。因此，本发明的范围应仅由所附权利要求书限制。

Claims (12)

  1. 一种目标检测方法，包括：
    分别获取摄像头和激光雷达同步采集到的图像和点云，并且从所述点云中提取位于所述摄像头的视场角内的点云作为第一点云；
    将所述第一点云投影至图像坐标系中得到所述图像坐标系下的第二点云；
    检测所述图像中包含的各目标得到各所述目标的各目标检测信息，各所述目标检测信息包括各所述目标的像素位置；
    根据各所述目标的所述像素位置确定各所述目标的目标框，根据位于各所述目标框内的所述第二点云确定与各所述目标对应的视锥点云；
    对各所述视锥点云进行坐标转换得到与各所述目标对应的目标视锥点云；
    提取各所述目标视锥点云中的第一目标点云，对各所述第一目标点云进行坐标转换得到与各所述目标对应的第二目标点云；
    对各所述第二目标点云进行拟合回归得到各所述目标的第一三维信息；
    根据各所述第一三维信息确定各所述目标在车辆坐标系下的第二三维信息，输出各所述目标检测信息和各所述第二三维信息。
  2. 根据权利要求1所述的方法，其中，所述将所述第一点云投影至图像坐标系中得到所述图像坐标系下的第二点云，包括：
    标定所述摄像头的内参数得到内参数矩阵；
    以所述摄像头为参考点联合标定所述摄像头和所述激光雷达得到所述激光雷达到所述摄像头的第一坐标转换矩阵；
    利用所述第一坐标转换矩阵对所述第一点云进行坐标转换得到摄像头坐标系下的第三点云；
    利用所述内参数矩阵对所述第三点云进行坐标投影得到所述图像坐标系下的所述第二点云。
  3. 根据权利要求2所述的方法，其中，所述根据位于各所述目标框内的所述第二点云确定与各所述目标对应的视锥点云包括：
    对位于各所述目标框内的所述第二点云进行随机采样，得到与各所述目标对应的第四点云；
    将与各所述第四点云对应的各所述第三点云确定为与各所述目标对应的所述视锥点云。
  4. 根据权利要求2所述的方法，其中，所述对各所述视锥点云进行坐标转 换得到与各所述目标对应的目标视锥点云包括：
    确定各所述目标的中心轴与所述摄像头坐标系的Z轴间的夹角，利用各所述夹角对各所述视锥点云进行坐标转换得到与各所述目标对应的第一视锥点云，其中，所述中心轴为所述目标的中心与所述摄像头坐标系的原点间的连线；
    计算各所述第一视锥点云的Y轴平均值，利用各所述Y轴平均值对各所述第一视锥点云进行坐标转换得到与各所述目标对应的所述目标视锥点云。
  5. 根据权利要求4所述的方法，其中，各所述第一三维信息包括各所述目标的中心点的坐标、长宽高及航向角，所述根据各所述第一三维信息确定各所述目标在车辆坐标系下的第二三维信息包括：
    以车辆的惯性测量装置为参考点标定所述摄像头的外参数得到外参数矩阵；
    对于每一所述目标，根据所述中心点的坐标、所述长宽高及所述航向角确定所述目标的各角点的坐标；
    利用所述第一目标点云的坐标平均值对所述中心点和各所述角点的坐标进行转换，得到第一中心点和各第一角点的坐标；
    利用所述第一视锥点云的Y轴平均值对所述第一中心点和各所述第一角点的坐标进行转换，得到第二中心点和各第二角点的坐标；
    利用所述目标的中心轴与所述摄像头坐标系的Z轴间的所述夹角对所述第二中心点和各所述第二角点的坐标进行转换，得到第三中心点和各第三角点的坐标；
    利用所述外参数矩阵对所述第三中心点和各所述第三角点的坐标进行转换，得到第四中心点和各第四角点的坐标；
    根据各所述第四角点的坐标计算所述目标的目标航向角，将所述第四中心点的坐标、所述长宽高及所述目标航向角确定为所述目标的所述第二三维信息；
    遍历各所述目标的所述第一三维信息，确定各所述目标在所述车辆坐标系下的所述第二三维信息。
  6. 根据权利要求1至5中任一项所述的方法，其中，所述提取各所述目标视锥点云中的第一目标点云包括：
    将各所述目标视锥点云分别输出至点云分割网络得到各所述目标视锥点云中各点属于前景的第一概率和属于背景的第二概率；
    对于各所述目标视锥点云，选取所述第一概率大于所述第二概率的所述点组成目标点云集合；
    对各所述目标点云集合进行随机采样得到各所述第一目标点云。
  7. 根据权利要求1至6中任一项所述的方法，其中，所述对各所述第一目标点云进行坐标转换得到与各所述目标对应的第二目标点云包括：
    计算各所述第一目标点云的坐标平均值；
    利用各所述第一目标点云的坐标平均值对各所述第一目标点云进行坐标转换，得到与各所述目标对应的所述第二目标点云。
  8. 根据权利要求1至7中任一项所述的方法，其中，所述检测所述图像中包含的各目标得到各所述目标的各目标检测信息包括：
    将所述图像输出至目标检测神经网络中以检测所述图像包含的各所述目标并得到各所述目标的各所述目标检测信息。
  9. 根据权利要求8所述的方法，其中，所述目标检测信息包括分类信息、像素位置及置信度中的至少一个。
  10. 根据权利要求1所述的方法，其中，所述对各所述第二目标点云进行拟合回归得到各所述目标的第一三维信息包括：
    将各所述第二目标点云输出至点云拟合回归网络中得到各所述目标的所述第一三维信息。
  11. 一种电子设备，包括：
    处理器；
    存储有计算机程序的存储器，
    其中，当所述计算机程序被所述处理器运行时，导致所述电子设备执行如权利要求1-10任一项所述的方法。
  12. 一种计算机存储介质，其中，所述存储介质中存储有至少一条指令、至少一段程序、代读码集或指令集，所述至少一条指令、至少一段程序、代码集或指令集由处理器加载并执行如权利要求1-10中任意一项所述的方法。
PCT/CN2021/111973 2021-03-01 2021-08-11 目标检测方法、电子介质和计算机存储介质 WO2022183685A1 (zh)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202110222540.4A CN112562093B (zh) 2021-03-01 2021-03-01 目标检测方法、电子介质和计算机存储介质
CN202110222540.4 2021-03-01

Publications (1)

Publication Number Publication Date
WO2022183685A1 true WO2022183685A1 (zh) 2022-09-09

Family

ID=75034853

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2021/111973 WO2022183685A1 (zh) 2021-03-01 2021-08-11 目标检测方法、电子介质和计算机存储介质

Country Status (2)

Country Link
CN (1) CN112562093B (zh)
WO (1) WO2022183685A1 (zh)

Families Citing this family (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112562093B (zh) * 2021-03-01 2021-05-18 湖北亿咖通科技有限公司 目标检测方法、电子介质和计算机存储介质
CN113160324B (zh) * 2021-03-31 2024-09-20 北京京东乾石科技有限公司 包围框生成方法、装置、电子设备和计算机可读介质
CN114331822A (zh) * 2021-12-29 2022-04-12 北京淳中科技股份有限公司 图像处理方法、装置和系统
CN115297315B (zh) * 2022-07-18 2024-07-26 北京城市网邻信息技术有限公司 用于环拍时拍摄中心点的矫正方法、装置及电子设备
CN116883496B (zh) * 2023-06-26 2024-03-12 小米汽车科技有限公司 交通元素的坐标重建方法、装置、电子设备及存储介质

Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106289071A (zh) * 2016-08-18 2017-01-04 温州大学 一种结构三维位移单目摄像测量方法
CN109784333A (zh) * 2019-01-22 2019-05-21 中国科学院自动化研究所 基于点云带权通道特征的三维目标检测方法及系统
CN110246159A (zh) * 2019-06-14 2019-09-17 湖南大学 基于视觉和雷达信息融合的3d目标运动分析方法
US10565787B1 (en) * 2017-01-27 2020-02-18 NHIAE Group, LLC Systems and methods for enhanced 3D modeling of a complex object
CN112562093A (zh) * 2021-03-01 2021-03-26 湖北亿咖通科技有限公司 目标检测方法、电子介质和计算机存储介质

Family Cites Families (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US11157014B2 (en) * 2016-12-29 2021-10-26 Tesla, Inc. Multi-channel sensor simulation for autonomous control systems
CN108564629A (zh) * 2018-03-23 2018-09-21 广州小鹏汽车科技有限公司 一种车载摄像头外部参数的标定方法及系统
CN108663677A (zh) * 2018-03-29 2018-10-16 上海智瞳通科技有限公司 一种多传感器深度融合提高目标检测能力的方法
CN110264416B (zh) * 2019-05-28 2020-09-29 深圳大学 稀疏点云分割方法及装置
CN111145174B (zh) * 2020-01-02 2022-08-09 南京邮电大学 基于图像语义特征进行点云筛选的3d目标检测方法
CN111951305B (zh) * 2020-08-20 2022-08-23 重庆邮电大学 一种基于视觉和激光雷达的目标检测和运动状态估计方法
CN112101128B (zh) * 2020-08-21 2021-06-22 东南大学 一种基于多传感器信息融合的无人方程式赛车感知规划方法
CN112257692B (zh) * 2020-12-22 2021-03-12 湖北亿咖通科技有限公司 一种行人目标的检测方法、电子设备及存储介质

Patent Citations (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN106289071A (zh) * 2016-08-18 2017-01-04 温州大学 一种结构三维位移单目摄像测量方法
US10565787B1 (en) * 2017-01-27 2020-02-18 NHIAE Group, LLC Systems and methods for enhanced 3D modeling of a complex object
CN109784333A (zh) * 2019-01-22 2019-05-21 中国科学院自动化研究所 基于点云带权通道特征的三维目标检测方法及系统
CN110246159A (zh) * 2019-06-14 2019-09-17 湖南大学 基于视觉和雷达信息融合的3d目标运动分析方法
CN112562093A (zh) * 2021-03-01 2021-03-26 湖北亿咖通科技有限公司 目标检测方法、电子介质和计算机存储介质

Also Published As

Publication number Publication date
CN112562093B (zh) 2021-05-18
CN112562093A (zh) 2021-03-26

Similar Documents

Publication Publication Date Title
WO2022183685A1 (zh) 目标检测方法、电子介质和计算机存储介质
CN112861653B (zh) 融合图像和点云信息的检测方法、系统、设备及存储介质
CN110146869B (zh) 确定坐标系转换参数的方法、装置、电子设备和存储介质
CN113657224B (zh) 车路协同中用于确定对象状态的方法、装置、设备
CN108921925B (zh) 基于激光雷达和视觉融合的语义点云生成方法和装置
CN110826499A (zh) 物体空间参数检测方法、装置、电子设备及存储介质
WO2021213432A1 (zh) 数据融合
CN113281780B (zh) 对图像数据进行标注的方法、装置及电子设备
CN113205604A (zh) 一种基于摄像机和激光雷达的可行区域检测方法
CN114463303B (zh) 一种基于双目相机和激光雷达融合的道路目标检测方法
US20160275359A1 (en) Information processing apparatus, information processing method, and computer readable medium storing a program
CN113160349A (zh) 点云标注方法、装置、存储介质及电子设备
KR102490521B1 (ko) 라이다 좌표계와 카메라 좌표계의 벡터 정합을 통한 자동 캘리브레이션 방법
CN115690469A (zh) 一种双目图像匹配方法、装置、设备和存储介质
CN115410167A (zh) 目标检测与语义分割方法、装置、设备及存储介质
CN115376109A (zh) 障碍物检测方法、障碍物检测装置以及存储介质
CN110197104B (zh) 基于车辆的测距方法及装置
CN114662587A (zh) 一种基于激光雷达的三维目标感知方法、装置及系统
WO2024183477A1 (zh) 一种鱼眼可行驶区域评测方法及装置
CN112364693B (zh) 基于双目视觉的障碍识别方法、装置、设备及存储介质
CN112036359B (zh) 一种车道线的拓扑信息获得方法、电子设备及存储介质
CN112598736A (zh) 一种基于地图构建的视觉定位方法及装置
CN112097742B (zh) 一种位姿确定方法及装置
CN116778262B (zh) 一种基于虚拟点云的三维目标检测方法和系统
CN113112551B (zh) 相机参数的确定方法、装置、路侧设备和云控平台

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 21928762

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 21928762

Country of ref document: EP

Kind code of ref document: A1