目标检测

使用卷积神经网络（CNN 或 ConvNet）执行分类、目标检测、迁移学习，创建自定义的检测器

目标检测是一种计算机视觉方法，用于定位图像或视频中的目标实例。目标检测算法通常利用机器学习或深度学习来生成有意义的结果。当查看图像或观看视频时，人类可以在瞬间识别并定位感兴趣的目标。目标检测旨在使用计算机复现这种智能。目标检测的最佳方法取决于您的应用和所要尝试解决的问题。

深度学习方法需要大量经过标注的训练图像，因此，推荐使用 GPU 来减少训练模型所需的时间。基于深度学习的目标检测方法使用卷积神经网络（即 CNN 或 ConvNet），比如 YOLO，或使用单次检测 (SSD) 方法。您可以训练自定义目标检测器，或通过迁移学习方法来使用预训练的目标检测器，这种方法使您能够从预训练网络开始，然后针对您的应用进行微调。卷积神经网络需要 Deep Learning Toolbox™。具有 CUDA^® 功能的 GPU 支持训练和预测。推荐使用 GPU，并且需要有 Parallel Computing Toolbox™。有关详细信息，请参阅Computer Vision Toolbox 预设项和MathWorks 产品中的并行计算支持 (Parallel Computing Toolbox)。

用于目标检测的机器学习方法包括聚合通道特征 (ACF)、使用有向梯度直方图 (HOG) 特征的支持向量机 (SVM) 分类以及用于人脸或上身检测的 Viola-Jones 算法。您可以选择从预训练的目标检测器开始，也可以根据您的应用创建自定义的目标检测器。

Labeled boats, neural network, and person detector

App

图像标注器	为计算机视觉应用标注图像
视频标注器	Label video for computer vision applications

函数

全部展开

检测目标

深度学习检测器

`rtmdetObjectDetector`	Detect objects using RTMDet object detector (自 R2024b 起)
`ssdObjectDetector`	Detect objects using SSD deep learning detector
`yolov2ObjectDetector`	Detect objects using YOLO v2 object detector
`yolov3ObjectDetector`	Detect objects using YOLO v3 object detector (自 R2021a 起)
`yolov4ObjectDetector`	Detect objects using YOLO v4 object detector (自 R2022a 起)
`yoloxObjectDetector`	Detect objects using YOLOX object detector (自 R2023b 起)
`peopleDetector`	Detect people using pretrained deep learning object detector (自 R2024b 起)
`faceDetector`	Detect faces using pretrained RetinaFace face detector (自 R2025a 起)

基于特征的检测器

`readAprilTag`	Detect and estimate pose for AprilTag in image
`readArucoMarker`	Detect and estimate pose for ArUco marker in image (自 R2024a 起)
`generateArucoMarker`	Generate ArUco marker images (自 R2024a 起)
`readBarcode`	Detect and decode 1-D or 2-D barcode in image
`acfObjectDetector`	Detect objects using aggregate channel features
`peopleDetectorACF`	Detect people using aggregate channel features
`vision.CascadeObjectDetector`	Detect objects using the Viola-Jones algorithm
`vision.ForegroundDetector`	Foreground detection using Gaussian mixture models
`vision.BlobAnalysis`	Properties of connected regions

使用点特征检测目标

`detectBRISKFeatures`	Detect BRISK features
`detectFASTFeatures`	Detect corners using FAST algorithm
`detectHarrisFeatures`	使用哈里斯-斯蒂芬斯算法检测角点
`detectKAZEFeatures`	Detect KAZE features
`detectMinEigenFeatures`	Detect corners using minimum eigenvalue algorithm
`detectMSERFeatures`	Detect MSER features
`detectORBFeatures`	Detect ORB keypoints
`detectSIFTFeatures`	Detect scale invariant feature transform (SIFT) features (自 R2021b 起)
`detectSURFFeatures`	检测 SURF 特征
`extractFeatures`	Extract interest point descriptors
`matchFeatures`	Find matching features

选择检测到的目标

`selectStrongestBbox`	Select strongest bounding boxes from overlapping clusters using nonmaximal suppression (NMS)
`selectStrongestBboxMulticlass`	Select strongest multiclass bounding boxes from overlapping clusters using nonmaximal suppression (NMS)

训练自定义目标检测器

加载训练数据

`boxLabelDatastore`	Datastore for bounding box label data
`groundTruth`	Ground truth label data
`imageDatastore`	图像数据的数据存储
`objectDetectorTrainingData`	Create training data for an object detector
`combine`	合并来自多个数据存储的数据

训练基于特征的目标检测器

`trainACFObjectDetector`	Train ACF object detector
`trainCascadeObjectDetector`	Train cascade object detector model
`trainImageCategoryClassifier`	Train an image category classifier

训练基于深度学习的目标检测器

`trainSSDObjectDetector`	Train SSD deep learning object detector
`trainYOLOv2ObjectDetector`	Train YOLO v2 object detector
`trainYOLOv3ObjectDetector`	Train YOLO v3 object detector (自 R2024a 起)
`trainYOLOv4ObjectDetector`	Train YOLO v4 object detector (自 R2022a 起)
`trainYOLOXObjectDetector`	Train YOLOX object detector (自 R2023b 起)

增强和预处理用于深度学习的训练数据

`balanceBoxLabels`	Balance bounding box labels for object detection
`bboxcrop`	Crop bounding boxes
`bboxerase`	Remove bounding boxes (自 R2021a 起)
`bboxresize`	Resize bounding boxes
`bboxwarp`	Apply geometric transformation to bounding boxes
`bbox2points`	Convert rectangle to corner points list
`blockLocationsWithROI`	Select image block locations that contain bounding box ROIs (自 R2025a 起)
`imwarp`	对图像应用几何变换
`imcrop`	裁剪图像
`imresize`	调整图像大小
`randomAffine2d`	Create randomized 2-D affine transformation
`centerCropWindow2d`	Create rectangular center cropping window
`randomWindow2d`	Randomly select rectangular region in image (自 R2021a 起)
`integralImage`	Calculate 2-D integral image

设计目标检测深度神经网络

R-CNN（基于卷积神经网络的区域）

`roiAlignLayer`	Non-quantized ROI pooling layer for Mask-CNN
`roiMaxPooling2dLayer`	Neural network layer used to output fixed-size feature maps for rectangular ROIs
`roialign`	Non-quantized ROI pooling of `dlarray` data (自 R2021b 起)

YOLO v2（you only look once 版本 2）

`yolov2TransformLayer`	Create transform layer for YOLO v2 object detection network
`spaceToDepthLayer`	Space to depth layer

焦点损失

focalCrossEntropy Compute focal cross-entropy loss

SSD（单次检测器）

ssdMergeLayer Create SSD merge layer for object detection

锚框

estimateAnchorBoxes Estimate anchor boxes for deep learning object detectors

可视化检测结果

`cuboid2img`	Project cuboids from 3-D world coordinates to 2-D image coordinates (自 R2022b 起)
`insertObjectAnnotation`	Annotate truecolor or grayscale image or video
`insertObjectMask`	Insert masks in image or video stream
`insertShape`	Insert shapes in image or video
`showShape`	Display shapes on image, video, or point cloud

评估预测结果

`evaluateObjectDetection`	Evaluate object detection data set against ground truth (自 R2023b 起)
`objectDetectionMetrics`	Object detection quality metrics (自 R2023b 起)
`mAPObjectDetectionMetric`	Mean average precision (mAP) metric for object detection (自 R2024a 起)
`bboxOverlapRatio`	Compute bounding box overlap ratio
`bboxPrecisionRecall`	Compute bounding box precision and recall against ground truth

模块

Deep Learning Object Detector

使用经过训练的深度学习目标检测器检测目标 (自 R2021b 起)

主题

快速入门

Get Started with Object Detection Using Deep Learning
Perform object detection using deep learning neural networks such as YOLOX, YOLO v4, and SSD.
Choose an Object Detector
Compare object detection deep learning models, such as YOLOX, YOLO v4, RTMDet, and SSD.
Local Feature Detection and Extraction
Learn the benefits and applications of local feature detection and extraction.
Get Started with Cascade Object Detector
Train a custom classifier.
Point Feature Types
Choose functions that return and accept points objects for several types of features.
Getting Started with OCR
Detect and recognize text in multiple languages, train OCR models to recognize custom text.
Image Classification with Bag of Visual Words
Use the Computer Vision Toolbox™ functions for image category classification by creating a bag of visual words.

用于目标检测和实例分割的训练数据

Get Started with the Image Labeler
Interactively label rectangular ROIs for object detection, pixels for semantic segmentation, polygons for instance segmentation, and scenes for image classification.
Get Started with the Video Labeler
Interactively label rectangular ROIs for object detection, pixels for semantic segmentation, polygons for instance segmentation, and scenes for image classification in a video or image sequence.
Datastores for Deep Learning (Deep Learning Toolbox)
Learn how to use datastores in deep learning applications.
Training Data for Object Detection and Semantic Segmentation
Create training data for object detection or semantic segmentation using the Image Labeler or Video Labeler.
Get Started with Image Preprocessing and Augmentation for Deep Learning
Preprocess data for deep learning applications with deterministic operations such as resizing, or augment training data with randomized operations such as random cropping.

深度学习快速入门

在 MATLAB 中进行深度学习 (Deep Learning Toolbox)
通过使用卷积神经网络进行分类和回归来探索 MATLAB^® 的深度学习能力，包括预训练网络和迁移学习，以及在 GPU、CPU、集群和云上进行训练。
预训练的深度神经网络 (Deep Learning Toolbox)
了解如何下载和使用预训练的卷积神经网络进行分类、迁移学习和特征提取。

精选示例

Detect Small Objects Using Tiled Training of YOLOX Network

Detect small objects in full-resolution images using tiled training of a you only look once version X (YOLOX) deep learning network.

自 R2024b 起
打开实时脚本

Object Detection in Large Satellite Imagery Using Deep Learning

Perform object detection on large satellite imagery using deep learning.

打开实时脚本

Object Detection Using YOLO v4 Deep Learning

Detect objects in images using you only look once version 4 (YOLO v4) deep learning network. In this example, you will

打开实时脚本

Multiclass Object Detection Using YOLO v2 Deep Learning

Train a YOLO v2 multiclass object detector and evaluate object detector performance across selected classes and overlap thresholds.

自 R2024b 起
打开实时脚本

Perform 6-DoF Pose Estimation for Bin Picking Using Deep Learning

Perform six degrees-of-freedom (6-DoF) pose estimation by estimating the 3-D position and orientation of machine parts in a bin using RGB-D images and a deep learning network.

打开实时脚本

Train Object Detectors in Experiment Manager

Use the Experiment Manager app to find optimal training options for object detectors.

打开脚本

Find Object in Cluttered Scene Using Image Point Features

Detect a particular object in a cluttered scene, given a reference image of the object.

打开脚本

Read Barcodes in Image

Detect, decode, and localize 1-D and 2-D barcodes in an image.

打开实时脚本

Detect Cars Using Gaussian Mixture Models

Detect and count cars in a video sequence using foreground detector based on Gaussian mixture models (GMMs).

打开脚本

Perform Instance Segmentation Using Mask R-CNN

Segment individual instances of people and cars using a multiclass mask region-based convolutional neural network (R-CNN).

打开实时脚本

Import Pretrained ONNX YOLO v2 Object Detector

Import pretrained YOLO v2 object detector from ONNX deep learning framework.

打开实时脚本

Export YOLO v2 Object Detector to ONNX

Export pretrained YOLO v2 object detector to ONNX deep learning framework.

打开实时脚本

Generate Code for Detecting Objects in Images by Using ACF Object Detector

Generate code from a MATLAB® function that detects objects in images by using an acfObjectDetector object. When you intend to generate code from your MATLAB function that uses an acfObjectDetector object, you must create the object outside of the MATLAB function. The example explains how to modify the MATLAB code in Train Stop Sign Detector Using ACF Object Detector to support code generation.

打开实时脚本

Code Generation for Object Detection by Using YOLO v2

Generate CUDA® code for object detection using YOLO v2.

打开实时脚本

Code Generation for Object Detection by Using Single Shot Multibox Detector

Generate CUDA code for an SSD network.

打开实时脚本

Code Generation for People Detection Using Deep Learning

Generate CUDA code for people detection

自 R2025a 起
打开实时脚本