默认情况下在 主机 上运行。在 RVC4 上,您可以使用
setRunOnHost(False) 将其卸载到设备。在 RVC2 上,仅支持主机端处理。在 DepthAI v3.6.0 中添加。如何放置
Python
Python
1import depthai as dai
2
3with dai.Pipeline() as pipeline:
4 pointCloud = pipeline.create(dai.node.PointCloud)输入和输出
passthroughDepth 转发用于计算给定 PointCloudData 输出的原始深度帧。当输入队列是非阻塞的并且您需要将深度帧与其相应的点云关联时非常有用。配置
initialConfig 设置,或者在运行时通过 inputConfig 发送。在流中发送新配置是安全的——当配置、校准或帧转换发生变化时,节点会自动重新初始化。有关详细说明和使用示例,请参阅 PointCloudConfig。Python
Python
1pc = pipeline.create(dai.node.PointCloud)
2pc.initialConfig.setOrganized(True)
3pc.initialConfig.setLengthUnit(dai.LengthUnit.METER)
4pc.initialConfig.setTargetCoordinateSystem(dai.CameraBoardSocket.CAM_A)
5pc.setNumFramesPool(8)坐标系
ImgTransformation 外参中读取 T_frame_to_ref 并应用它,因此输出点云位于参考(原点)相机坐标系中。在此基础上,您可以应用以下三种额外转换之一:1. 相机插座坐标系
CameraBoardSocket.CAM_A)。原点位于相机传感器处,旋转遵循相机的光学帧。节点使用设备校准自动计算外参转换。Python
1# 彩色相机 (CAM_A) 坐标系中的点云
2pc.initialConfig.setTargetCoordinateSystem(dai.CameraBoardSocket.CAM_A)2. 外壳坐标系
Python
1# VESA 安装坐标系中的点云
2pc.initialConfig.setTargetCoordinateSystem(dai.HousingCoordinateSystem.VESA_A)HousingCoordinateSystem.CAM_A–CAM_D— 原点与CameraBoardSocket.CAM_A–CAM_D相同。区别在于旋转:X-Y 平面与设备的前玻璃平行,而相机插座坐标系则旋转到相机的光学帧。HousingCoordinateSystem.FRONT_CAM_A–FRONT_CAM_D— 位置类似于HousingCoordinateSystem.CAM_A–CAM_D,但向前移动到前玻璃的前侧。HousingCoordinateSystem.VESA_A–VESA_J— 设备安装点。HousingCoordinateSystem.IMU— IMU 传感器原点。
3. 自定义变换矩阵
T_ref_to_custom — 从参考相机到您的自定义坐标系的变换。节点将其与帧外参组合以获得最终应用的变换:Python
1# 绕 Z 轴旋转 90°
2transform = [
3 [0.0, -1.0, 0.0, 0.0],
4 [1.0, 0.0, 0.0, 0.0],
5 [0.0, 0.0, 1.0, 0.0],
6 [0.0, 0.0, 0.0, 1.0],
7]
8pc.initialConfig.setTransformationMatrix(transform)T_frame_to_ref),将点带入参考相机帧。选项之间的交互
CAMERA_SOCKET 或 HOUSING — 在这些模式下,setTransformationMatrix 中的自定义变换矩阵将被忽略。自定义矩阵仅在坐标系类型为 DEFAULT 时使用(即未调用 setTargetCoordinateSystem 重载)。所有三种方法也可以直接在节点上调用(转发到 initialConfig):Python
1pc.setTargetCoordinateSystem(dai.CameraBoardSocket.CAM_A)
2# 或
3pc.setTargetCoordinateSystem(dai.HousingCoordinateSystem.VESA_A)用法
从立体深度生成点云
Python
Python
1import depthai as dai
2
3pipeline = dai.Pipeline()
4
5left = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_B)
6right = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_C)
7stereo = pipeline.create(dai.node.StereoDepth)
8left.requestOutput((640, 400)).link(stereo.left)
9right.requestOutput((640, 400)).link(stereo.right)
10
11pc = pipeline.create(dai.node.PointCloud)
12pc.initialConfig.setLengthUnit(dai.LengthUnit.METER)
13stereo.depth.link(pc.inputDepth)
14
15q = pc.outputPointCloud.createOutputQueue(maxSize=4, blocking=False)
16
17with pipeline:
18 pipeline.start()
19 while pipeline.isRunning():
20 pclData = q.get()
21 points = pclData.getPoints() # np.ndarray (N, 3) float32
22 print(f"Points: {len(points)}, Z=[{pclData.getMinZ():.2f}, {pclData.getMaxZ():.2f}]")彩色点云
Python
Python
1import depthai as dai
2
3pipeline = dai.Pipeline()
4
5left = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_B)
6right = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_C)
7color = pipeline.create(dai.node.Camera).build(dai.CameraBoardSocket.CAM_A)
8
9stereo = pipeline.create(dai.node.StereoDepth)
10left.requestFullResolutionOutput().link(stereo.left)
11right.requestFullResolutionOutput().link(stereo.right)
12
13colorOut = color.requestOutput((640, 400), type=dai.ImgFrame.Type.RGB888i,
14 resizeMode=dai.ImgResizeMode.CROP, enableUndistortion=True)
15
16pc = pipeline.create(dai.node.PointCloud)
17pc.initialConfig.setLengthUnit(dai.LengthUnit.METER)
18
19# 将深度与彩色相机对齐
20platform = pipeline.getDefaultDevice().getPlatform()
21if platform == dai.Platform.RVC4:
22 imageAlign = pipeline.create(dai.node.ImageAlign)
23 stereo.depth.link(imageAlign.input)
24 colorOut.link(imageAlign.inputAlignTo)
25 imageAlign.outputAligned.link(pc.inputDepth)
26else:
27 colorOut.link(stereo.inputAlignTo)
28 stereo.depth.link(pc.inputDepth)
29
30colorOut.link(pc.inputColor)
31
32q = pc.outputPointCloud.createOutputQueue(maxSize=4, blocking=False)
33
34with pipeline:
35 pipeline.start()
36 while pipeline.isRunning():
37 pcd = q.get()
38 if pcd.isColor():
39 xyz, rgba = pcd.getPointsRGB()
40 print(f"Points: {len(xyz)}, color=yes, Z=[{pcd.getMinZ():.2f}, {pcd.getMaxZ():.2f}]")示例
- PointCloud — 最小彩色点云示例。
- PointCloud Visualizer — 使用 Open3D 进行实时 3D 可视化。
- PointCloud Showcase — 演示过滤、组织、坐标变换、自定义矩阵和彩色模式。
参考
class
dai::node::PointCloud
variable
std::shared_ptr< PointCloudConfig > initialConfig
Initial config to use when computing the point cloud.
variable
Input inputConfig
Input PointCloudConfig message with ability to modify parameters in runtime. Default queue is non-blocking with size 4.
variable
Subnode< node::Sync > sync
variable
InputMap & syncInputs
variable
Input & inputDepth
Input message with depth data used to create the point cloud. Routed through the internal Sync subnode.
variable
Output outputPointCloud
variable
Output passthroughDepth
Passthrough depth from which the point cloud was calculated. Suitable for when input queue is set to non-blocking behavior.
function
PointCloud()function
PointCloud(std::unique_ptr< Properties > props)function
~PointCloud()function
Input & getColorInput()function
void setNumFramesPool(int numFramesPool)Specify number of frames in pool.
Parameters
- numFramesPool: How many frames should the pool have
function
void setRunOnHost(bool runOnHost)Specify whether to run on host or device By default, the node will run on host.
function
void useCPU()Use single-threaded CPU for processing
function
void useCPUMT(uint32_t numThreads)Use multi-threaded CPU for processing
function
void useGPU(uint32_t device)Use GPU for point cloud computation
Parameters
- device: GPU device index (default 0)
function
void setTargetCoordinateSystem(CameraBoardSocket targetCamera)Set target coordinate system to transform point cloud
Parameters
- targetCamera: Target camera socket
function
void setTargetCoordinateSystem(HousingCoordinateSystem housingCS)Set target coordinate system to housing coordinate system Point cloud will be transformed to this housing coordinate system
Parameters
- housingCS: Target housing coordinate system
function
void setTargetCoordinateSystem(CameraBoardSocket targetCamera, bool useSpecTranslation)Deprecated: use setTargetCoordinateSystem(targetCamera) instead.
function
void setTargetCoordinateSystem(HousingCoordinateSystem housingCS, bool useSpecTranslation)Deprecated: use setTargetCoordinateSystem(housingCS) instead.
function
bool runOnHost()Returns true or false whether the node should be run on host or not.
function
void buildInternal()Function called from within the create function to build the node. This function is useful for initialization, setting up inputs and outputs = stuff that cannot be perform in the constuctor.
class
dai::node::PointCloud::Impl
variable
LengthUnit targetLengthUnit
function
Impl()function
void setLogger(const std::shared_ptr<::spdlog::logger > & log)function
void computePointCloudDense(const uint8_t * depthData, std::vector< Point3f > & points)function
void computePointCloudDenseColored(const uint8_t * depthData, const uint8_t * colorData, std::vector< Point3fRGBA > & points)function
void applyTransformation(std::vector< PointT > & points)function
std::vector< PointT > filterValidPoints(const std::vector< PointT > & densePoints)function
void setLengthUnit(dai::LengthUnit lengthUnit)function
void useCPU()function
void useCPUMT(uint32_t numThreads)function
void useGPU(uint32_t device)function
void setIntrinsics(float fx, float fy, float cx, float cy, unsigned int width, unsigned int height)function
void setDistortion(CameraModel model, std::vector< float > coefficients)function
void setExtrinsics(const std::vector< std::vector< float >> & transformMatrix)function
void clearExtrinsics()需要帮助?
请访问讨论论坛以获取技术支持或您可能有的任何其他问题。