Video Classification

Perform video classification and activity recognition using deep learning

Classify the activity or action contained in a sequence of images from visual data sources, such as a video stream, into a set of categories using deep learning. Vision-based activity recognition involves predicting the action within a sequence of images, such as walking, swimming, or sitting, using a set of video frames. Activity recognition from video has many applications, such as human-computer interaction, anomaly detection, and surveillance. To learn more, see Getting Started with Video Classification Using Deep Learning.

Applications

Video Labeler	Label video for computer vision applications
Ground Truth Labeler	Label ground truth data for automated driving applications

Fonctions

développer tout

Extract Video Training Data

`groundTruth`	Ground truth label data
`writeVideoScenes`	Write video sequence to video file
`sceneTimeRanges`	Time ranges of scene labels from ground truth data

Load Video Training Data

`VideoReader`	Create object to read video files
`fileDatastore`	Datastore with custom file reader
`transform`	Transform datastore
`combine`	Combine data from multiple datastores
`folders2labels`	Get list of labels from folder names
`splitlabels`	Find indices to split labels according to specified proportions

Design Video Classifier

`inflated3dVideoClassifier`	Inflated-3D (I3D) video classifier. Requires Computer Vision Toolbox Model for Inflated-3D Video Classification
`slowFastVideoClassifier`	SlowFast video classifier. Requires Computer Vision Toolbox Model for SlowFast Video Classification
`r2plus1dVideoClassifier`	R(2+1)D video classifier. Requires Computer Vision Toolbox Model for R(2+1)D Video Classification

Train Video Classifier

`predict`	Compute video classifier predictions
`forward`	Compute video classifier outputs for training

Augment and Preprocess Training Data

`imwarp`	Apply geometric transformation to image
`imcrop`	Crop image
`imresize`	Resize image
`randomAffine2d`	Create randomized 2-D affine transformation
`centerCropWindow2d`	Create rectangular center cropping window
`randomCropWindow2d`	(Not recommended) Create randomized rectangular cropping window

Classify Video

`classifyVideoFile`	Classify a video file
`classifySequence`	Classify video sequence
`updateSequence`	Update video sequence for classification
`resetSequence`	Reset video sequence properties for streaming video classification

Visualize Classification Results

`vision.VideoPlayer`	Play video or display image
`vision.DeployableVideoPlayer`	Display video
`insertText`	Insert text in image or video

Rubriques

Getting Started with Video Classification Using Deep Learning
Video recognition and classification, analyze, classify, and track actions contained in visual data sources.

Exemples présentés

Activity Recognition from Video and Optical Flow Data Using Deep Learning

Train an inflated-3D (I3D) two-stream convolutional neural network for activity recognition using RGB and optical flow data from videos.

Ouvrir le live script

Gesture Recognition using Videos and Deep Learning

Train a SlowFast convolutional neural network for gesture recognition using RGB data from videos.

Ouvrir le live script