Computer Vision Toolbox

Design, simulate, calibrate, and deploy computer vision systems

Computer Vision Toolbox provides algorithms, apps, and AI models for designing, simulating, calibrating, and deploying computer vision systems. You can perform object detection and tracking, feature matching, and optical flow. You can automate camera calibration, including multi-sensor configurations. For 3D vision, the toolbox supports structure from motion, real-time SLAM, and novel view synthesis such as NeRF. Computer vision apps enable calibration and multi-user image and video labeling, including automation capabilities.

The toolbox provides AI techniques, including pretrained convolutional neural networks, vision transformers, and vision-language models. You can use these pretrained models for tasks such as image classification, object detection, segmentation, pose estimation, image captioning, visual question answering, and OCR, or further customize them through transfer learning.

You can generate code in C/C++ and HDL, and for GPU or NPU hardware targets (with MATLAB Coder, HDL Coder, GPU Coder, and hardware support packages). You can also build custom apps (with MATLAB Compiler).

Connect AI Agents to Computer Vision Toolbox

Bring domain-specific capabilities to your agentic AI workflow.

Learn more

Image and Video Ground Truth Labeling

Automate labeling for object detection, semantic segmentation, instance segmentation, and scene classification using the Video Labeler and Image Labeler apps.

Automatically Label Ground Truth Using Segment Anything Model

Documentation | Examples

Pedestrians, cars, and buses labeled using instance segmentation.

Deep Learning and Machine Learning

Train machine learning models and deep learning networks, or use pretrained networks, for object detection and segmentation. Evaluate the performance of these networks and deploy them by generating C/C++ or CUDA^® code.

Automated Data Labeling Using Vision-Language Models (6:11)

Documentation | Examples

Multiple fisheye images of a checkerboard used to calibrate a camera using the Camera Calibrator app.

Camera Calibration

Estimate intrinsic, extrinsic, and lens distortion parameters for monocular cameras and stereo camera pairs, using the Camera Calibrator app, Stereo Camera Calibrator app, or built-in functions.

Camera Calibration Using Custom Planar Calibration Patterns

Documentation | Examples

Visualization of camera-centric extrinsic parameters and calibration analysis.

Multi-Sensor Calibration

Estimate extrinsics between cameras, lidars, and IMUs. Perform robot hand-eye calibration using built-in functions. Use the Multi-Camera Calibrator app to calibrate multi-camera systems.

What is Multi-Sensor Calibration?

Documentation | Examples

Visual SLAM

Estimate camera position and orientation with respect to its surroundings while building a map of an unknown environment. Refine pose estimates using bundle adjustment and pose graph optimization.

Performant Monocular Visual-Inertial SLAM

Documentation | Examples

3D Reconstruction

Reconstruct the 3D structure of a scene from multiple 2D views using structure from motion (SfM). Generate photorealistic novel views using AI-driven techniques such as Neural Radiance Fields (NeRFs).

Structure from Motion from Multiple Views

Documentation | Examples

Two side-by-side images of a box and of the same box within a larger scene, with lines connecting individual matching features in the images.

Feature Detection, Extraction, and Matching

Detect, extract, and match features such as blobs, edges, and corners across multiple images. Use the matched features for registration, for object classification, or in complex workflows such as SLAM.

Object Detection in a Cluttered Scene Using Point Feature Matching

Documentation | Examples

Multiple pedestrians detected within the area of interest in a car dashcam video.

Multi-Object Tracking and Motion Estimation

Estimate motion and track multiple objects in video and image sequences.

Multi-Object Tracking with DeepSORT

Documentation | Examples

Camera path and point cloud map for deployed Visual SLAM algorithm.

Code Generation and Third-Party Support

Generate code from your computer vision algorithms for rapid prototyping, deployment, and verification. Integrate OpenCV-based projects and functions into MATLAB and Simulink.

Build and Deploy Visual SLAM Algorithm with ROS in MATLAB

Documentation | Examples

Product Resources:

Documentation Examples Videos Functions Blocks Requirements Release notes

Caterpillar Uses Big Data, Data Analytics, and Machine and Deep Learning to Build Ground-Truth for Training, Validation, and Deploying Classifiers

“We can access machine learning capabilities with a few lines of MATLAB code. Then, using code generation, engineers can deploy their trained classifier into the machine without manual intervention or delays in the process.”

View more customer stories

Computer Vision Toolbox FAQs

Computer Vision Toolbox provides algorithms and apps for designing and testing computer vision systems, including visual inspection, object detection and tracking, feature detection, extraction, and matching.

The toolbox provides pretrained convolutional neural networks (CNNs), vision transformers, and vision-language models for tasks like image classification, object detection, segmentation, pose estimation, captioning, visual question answering (VQA), and optical character recognition (OCR), as well as zero-shot models for vision tasks such as optical flow and 3D depth estimation.

Yes, the Video Labeler and Image Labeler apps enable team-based ground truth labeling with automation workflows for object detection, semantic segmentation, instance segmentation, and scene classification.

Computer Vision Toolbox automates calibration workflows for single, fisheye, stereo, and multi-camera configurations using the Camera Calibrator app, Stereo Camera Calibrator app, Multi-Camera Calibrator app or built-in functions.

Yes, the toolbox supports stereo vision, structure from motion, neural radiance fields (NeRF), and real-time visual SLAM for 3D vision applications.

You can generate code in C, C++, CUDA for GPU execution, and in hardware description languages (HDL) for rapid prototyping, deployment, and verification.