Choose SLAM Workflow Based on Sensor Data

You can use Computer Vision Toolbox™, Navigation Toolbox™, and Lidar Toolbox™ for Simultaneous Localization and Mapping (SLAM). SLAM is widely used in applications including automated driving, robotics, and unmanned aerial vehicles (UAV). To learn more about SLAM, see What is SLAM?.

Choose SLAM Workflow

To choose the right SLAM workflow for your application, consider what type of sensor data you are collecting. MATLAB^® support SLAM workflows that use images from a monocular or stereo camera system, or point cloud data including 2-D and 3-D lidar data.

Visual SLAM

There are several approaches to visual SLAM, each suited to different sensors and applications. To compute 3-D points, you need to measure the depth from the camera to points on the 3-D object. Monocular, stereo, and RGB-D sensors provide this depth information using different techniques.

• Monocular SLAM — Single camera to estimate motion over time. However, it cannot determine absolute scale, such as the true size or distance of objects. You can use the monovslam object to estimate a camera's trajectory and reconstruct a sparse 3-D map from a sequence of monocular images.

• Stereo visual SLAM — Stereo camera setup to derive depth from disparities between images, providing absolute scale. You can use the stereovslam object to estimate camera motion and build a scaled 3-D map using synchronized stereo image pairs.

• RGB-D visual SLAM — Depth-sensing camera captures both color and depth information, providing precise, per-pixel depth measurements. You can use the rgbdvslam object to track camera pose and construct a dense 3-D map by combining RGB and depth data.

• Visual-inertial SLAM — Fuses visual data with motion information from an inertial measurement unit (IMU), enhancing robustness and accuracy, particularly during rapid movements or in visually challenging environments. You can use vSLAM objects to fuse IMU measurements with mono, stereo and RGB-D data.

This table summarizes the key features available for visual SLAM.

Sensor Data	Features	Topics	Examples	Toolbox	Code Generation
Monocular images	Feature detection, extraction, and matching Triangulation and bundle adjustment Data management for key frames and map points Loop closure detection using bag of features Similarity pose graph optimization Fusion with IMU	Implement Visual SLAM in MATLAB Visual SLAM	Performant and deployable workflows: Performant and Deployable Monocular Visual SLAM Performant Monocular Visual-Inertial SLAM Develop Visual SLAM Algorithm Using Unreal Engine Simulation Modular workflows: Monocular Visual Simultaneous Localization and Mapping Monocular Visual-Inertial SLAM	Computer Vision Toolbox Navigation Toolbox	Computer Vision Toolbox C/C++ Code Generation Computer Vision Toolbox GPU Code Generation
Stereo images	Stereo image rectification Feature detection, extraction, and matching Reconstruction from disparity, triangulation, and bundle adjustment Data management for key frames and map points Loop closure detection using bag of features Pose graph optimization Fusion with IMU	Visual SLAM	Performant and deployable workflows: Performant and Deployable Stereo Visual SLAM with Fisheye Images Develop Visual SLAM Algorithm Using Unreal Engine Simulation Stereo Visual SLAM for UAV Navigation in 3D Simulation Modular workflows: Stereo Visual Simultaneous Localization and Mapping	Computer Vision Toolbox Navigation Toolbox	Computer Vision Toolbox C/C++ Code Generation Computer Vision Toolbox GPU Code Generation
RGB-D images	Feature detection, extraction, and matching Reconstruction from depth images, triangulation, and bundle adjustment Data management for key frames and map points Loop closure detection using bag of features Pose graph optimization Fusion with IMU	Visual SLAM	Performant and deployable workflows: Simulate RGB-D Visual SLAM System with Cosimulation in Gazebo and Simulink (ROS Toolbox) Modular workflows: Visual SLAM with RGB-D Camera	Computer Vision Toolbox Navigation Toolbox	Computer Vision Toolbox C/C++ Code Generation Computer Vision Toolbox GPU Code Generation

Point Cloud, 2-D, and 3-D Lidar SLAM

This table summarizes the key features available for point cloud data including 2-D and 3-D lidar SLAM.

Sensor Data	Features	Topics	Examples	Toolbox	Code Generation
2-D lidar scans	Occupancy map building Vehicle pose estimation Pose graph optimization SLAM algorithm tuning SLAM Map Builder app	Mapping (Navigation Toolbox) SLAM (Navigation Toolbox) Registration and SLAM (Lidar Toolbox)	Build Map from 2-D Lidar Scans Using SLAM (Lidar Toolbox) Implement Simultaneous Localization And Mapping (SLAM) with Lidar Scans (Navigation Toolbox)	Navigation Toolbox Lidar Toolbox	Navigation Toolbox C/C++ Code Generation (Navigation Toolbox) Navigation Toolbox GPU Code Generation (Navigation Toolbox) Lidar Toolbox C/C++ Code Generation (Lidar Toolbox)
Point cloud data	Point cloud processing Registration Data management for map building Loop closure detection with global features Pose graph optimization Localization in a known map	Implement Point Cloud SLAM in MATLAB	Build a Map from Lidar Data Using SLAM Design Lidar SLAM Algorithm Using Unreal Engine Simulation Environment Build Occupancy Map from 3-D Lidar Data Using SLAM (Automated Driving Toolbox)	Computer Vision Toolbox	Computer Vision Toolbox C/C++ Code Generation Computer Vision Toolbox GPU Code Generation
3-D lidar scans	Feature-based: Registration Loop closure detection Localization in a known map	Registration and SLAM (Lidar Toolbox)	Aerial Lidar SLAM Using FPFH Descriptors (Lidar Toolbox) Build Map and Localize Using Segment Matching (Lidar Toolbox) Feature-Based Map Building from Lidar Data (Lidar Toolbox) Build a Map with Lidar Odometry and Mapping (LOAM) Using Unreal Engine Simulation (Lidar Toolbox)	Lidar Toolbox	Lidar Toolbox C/C++ Code Generation (Lidar Toolbox)