DepthAI v2 has been superseded by DepthAI v3. You are viewing legacy documentation.
此页面由 AI 自动翻译。查看英文原版
DepthAI 教程
DepthAI API 参考
最后更新

本页目录

  • 演示
  • 深度滤波器
  • 类似示例:
  • 设置
  • 源代码
  • 流水线
Claude CodeCodexCursor
Luxonis 智能体工具包为你常用的编程智能体加载 Luxonis 上下文与技能,更快地基于 OAK 开发应用。
安装工具包

深度后处理

最后更新
本示例展示了如何在设备本身运行深度后处理滤波器,以减少噪声、平滑深度图并整体提高深度图质量。后处理可以添加到 StereoDepth 节点。

演示

software/depthai/examples/depth_comparison.webp

深度滤波器

中值滤波

这是一种非边缘保留的中值滤波器,可用于减少噪声并平滑深度图。中值滤波器在硬件中实现,因此它是最快的滤波器。

散斑滤波

散斑滤波器用于减少散斑噪声。散斑噪声是相邻视差/深度像素之间方差很大的区域,散斑滤波器试图过滤此区域。
class

depthai.StereoDepthConfig.PostProcessing.SpeckleFilter

method
property
differenceThreshold
Maximum difference between neighbor disparity pixels to put them into the same blob. Units in disparity integer levels.
method
property
enable
Whether to enable or disable the filter.
method
property
speckleRange
Speckle search range.
method

时域滤波

时域滤波器旨在通过根据先前帧操作每个像素值来提高深度数据的持久性。该滤波器对数据执行单次通过,调整深度值,同时更新跟踪历史记录。在像素数据缺失或无效的情况下,滤波器使用用户定义的持久性模式来决定是否应使用存储的数据纠正缺失值。请注意,由于其对历史数据的依赖,该滤波器可能会引入可见的模糊/涂抹伪影,因此最适合静态场景。
class

depthai.StereoDepthConfig.PostProcessing.TemporalFilter

class
PersistencyMode
Persistency algorithm type.  Members:    PERSISTENCY_OFF :     VALID_8_OUT_OF_8 :     VALID_2_IN_LAST_3 :     VALID_2_IN_LAST_4 :     VALID_2_OUT_OF_8 :     VALID_1_IN_LAST_2 :     VALID_1_IN_LAST_5 :     VALID_1_IN_LAST_8 :     PERSISTENCY_INDEFINITELY : 
method
property
alpha
The Alpha factor in an exponential moving average with Alpha=1 - no filter. Alpha = 0 - infinite filter. Determines the extent of the temporal history that should be averaged.
method
property
delta
Step-size boundary. Establishes the threshold used to preserve surfaces (edges). If the disparity value between neighboring pixels exceed the disparity threshold set by this delta parameter, then filtering will be temporarily disabled. Default value 0 means auto: 3 disparity integer levels. In case of subpixel mode it's 3*number of subpixel levels.
method
property
enable
Whether to enable or disable the filter.
method
property
persistencyMode
Persistency mode. If the current disparity/depth value is invalid, it will be replaced by an older value, based on persistency mode.
method
class

depthai.StereoDepthConfig.PostProcessing.TemporalFilter.PersistencyMode

variable
variable
variable
variable
variable
variable
variable
variable
variable
variable
method
method
method
method
method
method
method
method
method
method
property
property

空间滤波

空间边缘保留滤波器将用有效的相邻深度像素填充无效的深度像素。它执行一系列一维水平和垂直通过或迭代,以增强重建数据的平滑度。它基于这篇研究论文。
class

depthai.StereoDepthConfig.PostProcessing.SpatialFilter

method
property
alpha
The Alpha factor in an exponential moving average with Alpha=1 - no filter. Alpha = 0 - infinite filter. Determines the amount of smoothing.
method
property
delta
Step-size boundary. Establishes the threshold used to preserve "edges". If the disparity value between neighboring pixels exceed the disparity threshold set by this delta parameter, then filtering will be temporarily disabled. Default value 0 means auto: 3 disparity integer levels. In case of subpixel mode it's 3*number of subpixel levels.
method
property
enable
Whether to enable or disable the filter.
method
property
holeFillingRadius
An in-place heuristic symmetric hole-filling mode applied horizontally during the filter passes. Intended to rectify minor artefacts with minimal performance impact. Search radius for hole filling.
method
property
numIterations
Number of iterations over the image in both horizontal and vertical direction.
method

阈值滤波

阈值滤波器过滤掉所有超出配置的最小/最大阈值范围的视差/深度像素。
Python
1class ThresholdFilter:
2'''
3    空间边缘保留滤波器将用有效的相邻深度像素填充无效的深度像素。
4    它执行一系列一维水平和垂直通过或迭代,以增强重建数据的平滑度。它基于[这篇研究论文](https://www.inf.ufrgs.br/~eslgastal/DomainTransform/)。
5'''

类似示例:

设置

请运行安装脚本以下载所有必需的依赖项。请注意,此脚本必须在git上下文中运行,因此您必须首先下载depthai-python存储库,然后运行该脚本
Command Line
1git clone https://github.com/luxonis/depthai-python.git
2cd depthai-python/examples
3python3 install_requirements.py
有关更多信息,请遵循安装指南。

源代码

Python

Python
GitHub
1#!/usr/bin/env python3
2
3import cv2
4import depthai as dai
5import numpy as np
6
7# Closer-in minimum depth, disparity range is doubled (from 95 to 190):
8extended_disparity = False
9# Better accuracy for longer distance, fractional disparity 32-levels:
10subpixel = False
11# Better handling for occlusions:
12lr_check = True
13
14# Create pipeline
15pipeline = dai.Pipeline()
16
17# Define sources and outputs
18monoLeft = pipeline.create(dai.node.MonoCamera)
19monoRight = pipeline.create(dai.node.MonoCamera)
20depth = pipeline.create(dai.node.StereoDepth)
21xout = pipeline.create(dai.node.XLinkOut)
22
23xout.setStreamName("disparity")
24
25# Properties
26monoLeft.setResolution(dai.MonoCameraProperties.SensorResolution.THE_400_P)
27monoLeft.setCamera("left")
28monoRight.setResolution(dai.MonoCameraProperties.SensorResolution.THE_400_P)
29monoRight.setCamera("right")
30
31# Create a node that will produce the depth map (using disparity output as it's easier to visualize depth this way)
32depth.setDefaultProfilePreset(dai.node.StereoDepth.PresetMode.HIGH_DENSITY)
33# Options: MEDIAN_OFF, KERNEL_3x3, KERNEL_5x5, KERNEL_7x7 (default)
34depth.initialConfig.setMedianFilter(dai.MedianFilter.KERNEL_7x7)
35depth.setLeftRightCheck(lr_check)
36depth.setExtendedDisparity(extended_disparity)
37depth.setSubpixel(subpixel)
38
39config = depth.initialConfig.get()
40config.postProcessing.speckleFilter.enable = False
41config.postProcessing.speckleFilter.speckleRange = 50
42config.postProcessing.temporalFilter.enable = True
43config.postProcessing.spatialFilter.enable = True
44config.postProcessing.spatialFilter.holeFillingRadius = 2
45config.postProcessing.spatialFilter.numIterations = 1
46config.postProcessing.thresholdFilter.minRange = 400
47config.postProcessing.thresholdFilter.maxRange = 15000
48config.postProcessing.decimationFilter.decimationFactor = 1
49depth.initialConfig.set(config)
50
51# Linking
52monoLeft.out.link(depth.left)
53monoRight.out.link(depth.right)
54depth.disparity.link(xout.input)
55
56# Connect to device and start pipeline
57with dai.Device(pipeline) as device:
58
59    # Output queue will be used to get the disparity frames from the outputs defined above
60    q = device.getOutputQueue(name="disparity", maxSize=4, blocking=False)
61
62    while True:
63        inDisparity = q.get()  # blocking call, will wait until a new data has arrived
64        frame = inDisparity.getFrame()
65        # Normalization for better visualization
66        frame = (frame * (255 / depth.initialConfig.getMaxDisparity())).astype(np.uint8)
67
68        cv2.imshow("disparity", frame)
69
70        # Available color maps: https://docs.opencv.org/3.4/d3/d50/group__imgproc__colormap.html
71        frame = cv2.applyColorMap(frame, cv2.COLORMAP_JET)
72        cv2.imshow("disparity_color", frame)
73
74        if cv2.waitKey(1) == ord('q'):
75            break

C++

1#include <iostream>
2
3// Inludes common necessary includes for development using depthai library
4#include "depthai/depthai.hpp"
5
6// Closer-in minimum depth, disparity range is doubled (from 95 to 190):
7static std::atomic<bool> extended_disparity{false};
8// Better accuracy for longer distance, fractional disparity 32-levels:
9static std::atomic<bool> subpixel{false};
10// Better handling for occlusions:
11static std::atomic<bool> lr_check{true};
12
13int main() {
14    // Create pipeline
15    dai::Pipeline pipeline;
16
17    // Define sources and outputs
18    auto monoLeft = pipeline.create<dai::node::MonoCamera>();
19    auto monoRight = pipeline.create<dai::node::MonoCamera>();
20    auto depth = pipeline.create<dai::node::StereoDepth>();
21    auto xout = pipeline.create<dai::node::XLinkOut>();
22
23    xout->setStreamName("disparity");
24
25    // Properties
26    monoLeft->setResolution(dai::MonoCameraProperties::SensorResolution::THE_400_P);
27    monoLeft->setCamera("left");
28    monoRight->setResolution(dai::MonoCameraProperties::SensorResolution::THE_400_P);
29    monoRight->setCamera("right");
30
31    // Create a node that will produce the depth map (using disparity output as it's easier to visualize depth this way)
32    depth->setDefaultProfilePreset(dai::node::StereoDepth::PresetMode::HIGH_DENSITY);
33    // Options: MEDIAN_OFF, KERNEL_3x3, KERNEL_5x5, KERNEL_7x7 (default)
34    depth->initialConfig.setMedianFilter(dai::MedianFilter::KERNEL_7x7);
35    depth->setLeftRightCheck(lr_check);
36    depth->setExtendedDisparity(extended_disparity);
37    depth->setSubpixel(subpixel);
38    auto config = depth->initialConfig.get();
39    config.postProcessing.speckleFilter.enable = false;
40    config.postProcessing.speckleFilter.speckleRange = 50;
41    config.postProcessing.temporalFilter.enable = true;
42    config.postProcessing.spatialFilter.enable = true;
43    config.postProcessing.spatialFilter.holeFillingRadius = 2;
44    config.postProcessing.spatialFilter.numIterations = 1;
45    config.postProcessing.thresholdFilter.minRange = 400;
46    config.postProcessing.thresholdFilter.maxRange = 15000;
47    config.postProcessing.decimationFilter.decimationFactor = 1;
48    depth->initialConfig.set(config);
49
50    // Linking
51    monoLeft->out.link(depth->left);
52    monoRight->out.link(depth->right);
53    depth->disparity.link(xout->input);
54
55    // Connect to device and start pipeline
56    dai::Device device(pipeline);
57
58    // Output queue will be used to get the disparity frames from the outputs defined above
59    auto q = device.getOutputQueue("disparity", 4, false);
60
61    while(true) {
62        auto inDepth = q->get<dai::ImgFrame>();
63        auto frame = inDepth->getFrame();
64        // Normalization for better visualization
65        frame.convertTo(frame, CV_8UC1, 255 / depth->initialConfig.getMaxDisparity());
66
67        cv::imshow("disparity", frame);
68
69        // Available color maps: https://docs.opencv.org/3.4/d3/d50/group__imgproc__colormap.html
70        cv::applyColorMap(frame, frame, cv::COLORMAP_JET);
71        cv::imshow("disparity_color", frame);
72
73        int key = cv::waitKey(1);
74        if(key == 'q' || key == 'Q') {
75            return 0;
76        }
77    }
78    return 0;
79}

流水线

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。