视觉 SLAM

使用单目、RGB-D 或立体相机进行实时视觉定位和地图构建 (vSLAM)，支持惯性传感器融合和部署

视觉同步定位与地图构建 (vSLAM) 是使用视觉输入估计相机位置和姿态，并同时进行环境地图构建的过程。Computer Vision Toolbox™ 支持单目、RGB-D 和立体相机的 vSLAM 工作流，并可选择惯性传感器融合以提高精度。这些功能对于机器人、增强现实和自主导航等应用至关重要。有关选择 vSLAM 工作流的指南，请参阅Choose SLAM Workflow Based on Sensor Data。

每个视觉 SLAM 对象（monovslam、rgbdvslam 和 stereovslam）都提供了即用型工具，可添加帧、跟踪关键帧、计算三维地图点、估计相机位姿、执行闭环检测以及在整个相机轨迹中可视化数据。您还可以通过使用 compareTrajectories 函数将估计的相机轨迹与真实值进行比较，以评估 vSLAM 算法的性能。工具箱还提供了构建您自己的视觉 SLAM 流程的功能。

您可以使用该工具箱执行 vSLAM 算法的代码生成和部署。有关详细信息，请参阅Build and Deploy Visual SLAM Algorithm with ROS in MATLAB和Performant and Deployable Monocular Visual SLAM。

函数

全部展开

单目视觉 SLAM

`monovslam`	Visual simultaneous localization and mapping (vSLAM) and visual-inertial sensor fusion with monocular camera (自 R2023b 起)
`addFrame`	Add image frame to visual SLAM object (自 R2023b 起)
`hasNewKeyFrame`	Check if new key frame added in visual SLAM object (自 R2023b 起)
`checkStatus`	Check status of visual SLAM object (自 R2023b 起)
`isDone`	End-of-file status (logical)
`mapPoints`	Build 3-D map of world points (自 R2023b 起)
`poses`	Absolute camera poses of key frames (自 R2023b 起)
`plot`	Plot 3-D map points and estimated camera trajectory in visual SLAM (自 R2023b 起)
`reset`	Reset visual SLAM object (自 R2023b 起)

RGB-D 视觉 SLAM

`rgbdvslam`	Feature-based visual simultaneous localization and mapping (vSLAM) and visual-inertial sensor fusion with RGB-D camera (自 R2024a 起)
`addFrame`	Add pair of color and depth images to RGB-D visual SLAM object (自 R2024a 起)
`hasNewKeyFrame`	Check if new key frame added in RGB-D visual SLAM object (自 R2024a 起)
`checkStatus`	Check status of visual RGB-D SLAM object (自 R2024a 起)
`isDone`	End-of-processing status for RGB-D visual SLAM object (自 R2024a 起)
`mapPoints`	Build 3-D map of world points from RGB-D vSLAM object (自 R2024a 起)
`poses`	Absolute camera poses of RGB-D vSLAM key frames (自 R2024a 起)
`plot`	Plot 3-D map points and estimated camera trajectory in RGB-D visual SLAM (自 R2024a 起)
`reset`	Reset RGB-D visual SLAM object (自 R2024a 起)

立体视觉 SLAM

`stereovslam`	Feature-based visual simultaneous localization and mapping (vSLAM) and visual-inertial sensor fusion with stereo camera (自 R2024a 起)
`addFrame`	Add pair of color and depth images to stereo visual SLAM object (自 R2024a 起)
`hasNewKeyFrame`	Check if new key frame added in stereo visual SLAM object (自 R2024a 起)
`checkStatus`	Check status of stereo visual SLAM object (自 R2024a 起)
`isDone`	End-of-processing status for stereo visual SLAM object (自 R2024a 起)
`mapPoints`	Build 3-D map of world points from stereo vSLAM object (自 R2024a 起)
`poses`	Absolute camera poses of stereo key frames (自 R2024a 起)
`plot`	Plot 3-D map points and estimated camera trajectory in stereo visual SLAM (自 R2024a 起)
`reset`	Reset stereo visual SLAM object (自 R2024a 起)

评估结果

`compareTrajectories`	Compare estimated trajectory against ground truth (自 R2024b 起)
`trajectoryErrorMetrics`	Store accuracy metrics for trajectories (自 R2024b 起)

可视化结果

`imshow`	显示图像
`showMatchedFeatures`	显示相应的特征点
`plot`	Plot image view set views and connections
`plotCamera`	Plot camera in 3-D coordinates
`pcshow`	Plot 3-D point cloud
`pcplayer`	Visualize streaming 3-D point cloud data

构建您自己的视觉 SLAM 流程

检测、提取和匹配特征

`detectSURFFeatures`	检测 SURF 特征
`detectORBFeatures`	检测 ORB 关键点
`extractFeatures`	提取兴趣点描述符
`matchFeatures`	查找匹配的特征
`matchFeaturesInRadius`	Find matching features within specified radius

重建三维结构

`triangulate`	立体图像中未去畸变的匹配点的三维位置
`img2world2d`	Determine world coordinates of image points (自 R2022b 起)
`world2img`	将世界点投影到图像中 (自 R2022b 起)

估计运动

`estgeotform2d`	Estimate 2-D geometric transformation from matching point pairs (自 R2022b 起)
`estgeotform3d`	Estimate 3-D geometric transformation from matching point pairs (自 R2022b 起)
`estimateFundamentalMatrix`	Estimate fundamental matrix from corresponding points in stereo images
`estworldpose`	Estimate camera pose from 3-D to 2-D point correspondences (自 R2022b 起)
`findWorldPointsInView`	Find world points observed in view
`findWorldPointsInTracks`	Find world points that correspond to point tracks
`estrelpose`	Calculate relative rotation and translation between camera poses (自 R2022b 起)

优化运动和三维结构

`optimizePoses`	Optimize absolute poses using relative pose constraints
`createPoseGraph`	Create pose graph
`bundleAdjustment`	Adjust collection of 3-D points and camera poses
`bundleAdjustmentMotion`	Adjust collection of 3-D points and camera poses using motion-only bundle adjustment
`bundleAdjustmentStructure`	Refine 3-D points using structure-only bundle adjustment

闭环检测

`bagOfFeatures`	Bag of visual words object
`bagOfFeaturesDBoW`	Bag of visual words using DBoW2 library (自 R2024b 起)
`dbowLoopDetector`	Detect loop closure using visual features (自 R2024b 起)
`indexImages`	Create image search index
`invertedImageIndex`	Search index that maps visual words to images

管理数据

`imageviewset`	Manage data for structure-from-motion, visual odometry, and visual SLAM
`worldpointset`	Manage 3-D to 2-D point correspondences

变换

`se3`	SE(3) homogeneous transformation (自 R2026a 起)
`so3`	SO(3) rotation (自 R2026a 起)

主题

即用型视觉 SLAM 函数

Performant and Deployable Monocular Visual SLAM
Use visual inputs from a camera to perform vSLAM and generate multi-threaded C/C++ code.
Performant Monocular Visual-Inertial SLAM
Use visual inputs from a camera and positional data from an IMU to perform viSLAM in real time. (自 R2025a 起)
Choose SLAM Workflow Based on Sensor Data
Choose the right simultaneous localization and mapping (SLAM) workflow and find topics, examples, and supported features.
How to Improve Accuracy in Visual SLAM
Tips to improve the accuracy, robustness, and efficiency of your visual SLAM system.

构建您自己的视觉 SLAM 流程

Monocular Visual Simultaneous Localization and Mapping
Visual simultaneous localization and mapping (vSLAM).
Monocular Visual-Inertial SLAM
Perform SLAM by combining images captured by a monocular camera with measurements from an IMU sensor.
Stereo Visual Simultaneous Localization and Mapping
Process image data from a stereo camera to build a map of an outdoor environment and estimate the trajectory of the camera.

精选示例

Performant and Deployable Monocular Visual SLAM

Use visual inputs from a camera to perform vSLAM and generate multi-threaded C/C++ code.

打开实时脚本

Performant Monocular Visual-Inertial SLAM

Use visual inputs from a camera and positional data from an IMU to perform viSLAM in real time.

自 R2025a 起
打开实时脚本

Performant and Deployable Stereo Visual SLAM with Fisheye Images

Use fisheye image data from a stereo camera to perform VSLAM and generate multi-threaded C/C++ code.

打开实时脚本

Build and Deploy Visual SLAM Algorithm with ROS in MATLAB

Implement and generate C ++ code for a vSLAM algorithm that estimates poses for the TUM RGB-D Benchmark and deploy as an ROS node to a remote device.

打开实时脚本

Simulate RGB-D Visual SLAM System with Cosimulation in Gazebo and Simulink

Simulates an RGB-D visual simultaneous localization and mapping (SLAM) system to estimate the camera poses using data from a mobile robot in Gazebo.

(ROS Toolbox)

自 R2024b 起

Stereo Visual SLAM for UAV Navigation in 3D Simulation

Develop a visual SLAM algorithm for a UAV equipped with a stereo camera.

打开实时脚本

Monocular Visual Odometry

Determine location and orientation of a camera by analyzing a sequence of images.

打开实时脚本

Develop Visual SLAM Algorithm Using Unreal Engine Simulation

Develop a visual simultaneous localization and mapping (SLAM) algorithm using image data from the Unreal Engine^® simulation environment.

(Automated Driving Toolbox)

全新

Estimate Camera-to-IMU Transformation Using Extrinsic Calibration

Estimate SE(3) transformation to define spatial relationship between camera and IMU.

自 R2026a 起
打开实时脚本

Visual Localization in a Parking Lot

Develop a visual localization system using synthetic image data from the Unreal Engine® simulation environment.

打开实时脚本