Human Activity Recognition Datasets for AI Training
License human-centered video and image data from Wavebreak Media's wholly owned archive for action recognition, gesture recognition, human-object interaction, and evaluation, or commission custom production for defined actions and capture conditions.
Datasets can be curated by activity class, sequence duration, temporal boundaries, participant count, object interaction, environment, viewpoint, execution variation, annotations, usage rights, and delivery format.
Featured Datasets
Explore sample licensed datasets across image, video, audio, and text collections.
Human Activity Data by Task
Configure human activity recognition (HAR) and action recognition datasets from archive-curated or custom-produced video around the required actions, sequence structure, variation, and labels. For broader visual tasks, review our computer vision datasets.
Video Action Classification
Trimmed clips or ordered frames showing defined actions, with class IDs, inclusion rules, duration criteria, and controlled dataset splits.
Temporal Action Localization
Continuous or multi-action video with action start and end times, segment labels, background intervals, and multi-label rules where commissioned.
Gesture Recognition
Short video or image sequences covering defined hand, arm, body, and communicative gestures across relevant viewpoints and execution styles.
Human-Object Interaction
People using, carrying, preparing, operating, or exchanging objects, organized by action-object pair, object category, participant role, and context where specified.
Multi-Person Activity Recognition
Conversation, collaboration, group exercise, family activity, and other interpersonal actions with participant and interaction labels where required. The People and Human Interaction Images Dataset provides related still-image and frame-level coverage.
Evaluation and Class Expansion
Held-out actions, underrepresented classes, hard negatives, and difficult temporal or visual conditions selected against defined evaluation criteria.
Model Evaluation Datasets
Activities and Environments Available
Wavebreak Media can assess existing footage and custom production across the following activity groups. Availability is confirmed against the required action, participant, environment, and sequence depth.
Specialized procedures, controlled scripts, exact object interactions, repeated viewpoints, and balanced participant coverage may require custom production.
For participant variation in frame-level classification and evaluation, review the Diverse People Images Dataset.
Daily Living and Household Routines
Preparing or eating food, cleaning, organizing, relaxing, communicating, and completing everyday activities in home and lifestyle settings. The Everyday Moments and Daily Life Images Dataset provides complementary still-image coverage.
Workplace Tasks and Tool Use
Computer use, meetings, manual tasks, equipment handling, collaboration, and routines across office and professional environments.
Fitness, Exercise, and Sports
Walking, running, stretching, strength training, gym activity, exercise, and sports movements shown through complete action sequences.
Product and Device Interaction
Using phones, computers, appliances, consumer products, tools, and equipment with the person, object, action, and setting visible where required.
Retail and Shopping Activities
Browsing shelves, selecting products, using carts or baskets, comparing items, queueing, paying, and retail interaction.
Mobility, Gestures, and Social Interaction
Standing, sitting, walking, carrying, pointing, greeting, conversing, cooperating, and participating in individual or group activities. The Milestone Celebrations Images Dataset adds group events and socially expressive contexts.
Temporal and Visual Coverage
Projects can combine video datasets for AI training with extracted frames or image datasets for AI training when both temporal and frame-level representations are required.
- Complete action sequences, including preparation, execution, and completion
- Action duration, speed, repetition, transition, and execution-style variation
- Camera viewpoint, orientation, distance, framing, movement, and resolution
- Single-person, multi-person, and person-object interaction coverage
- Indoor and outdoor settings, lighting conditions, backgrounds, and scene contexts
- Occlusion, clutter, wardrobe variation, body-position changes, and object-state changes
Labels, Annotations, and Dataset Splits
Selected human activity assets can be prepared with:
- Activity structure - stable asset and sequence IDs, hierarchical action classes, object relationships, participant counts, and multi-label rules.
- Temporal and spatial annotations - clip labels, timestamps, action segments, person or object boxes, tracks, keypoints, or pose data where commissioned.
- Context and technical metadata - scene, viewpoint, duration, frame rate, resolution, camera movement, captions, and specified descriptive fields.
- Split and delivery controls - actor-, shoot-, location-, and sequence-aware grouping, near-duplicate rules, manifests, media formats, checksums where required, and secure dataset delivery.
Human Activity Dataset Options
Existing footage, archive curation, and custom production can be used separately or combined under one dataset specification.
Ready-Made Human Motion Data
Review the Human Motion Dataset for current clip volume, activity coverage, specifications, sample assets, metadata, and licensing options.
View Human Motion DatasetArchive-Curated Activity Datasets
Build a project-specific licensed corpus from eligible archive footage when existing actions, environments, and sequence coverage match the brief.
Browse Dataset LibraryCustom Human Activity Production
Commission controlled capture for missing actions, participants, settings, viewpoints, repetitions, or hard negatives, with requirements defined before production.
Custom Dataset CreationWhy Wavebreak Media for Human Activity Recognition Data?
Since 2005, Wavebreak Media has produced and managed an extensive image archive and more than one million wholly owned video assets. Its human-centered footage covers everyday routines, exercise, work, retail, travel, product use, gestures, and social interaction.
Projects can combine archive curation with controlled capture, project-defined annotations, quality checks, and licensing for the intended AI use. Available provenance and applicable releases can be reviewed through the dataset licensing and compliance process.
Frequently Asked Questions (FAQ)
Not always. Existing assets may include descriptions and technical metadata. Action labels, timestamps, tracks, boxes, keypoints, or other annotations are confirmed or commissioned per project.
Yes, where source footage supports the sequence. Complete. Complete clips preserve temporal information; extracted frames support frame-level classification or annotation.
Yes, when action boundaries are annotated. Classification labels a clip; temporal detection also identifies when actions start and end within longer video.
Yes, when commissioned. Projects can specify boxes, tracks, action-object relationships, keypoints, pose data, participant roles, and temporal segments.
Yes. Existing footage can cover suitable activities; custom production fills gaps in actions, participants, objects, viewpoints, environments, or class balance.
Licensing is finalized for selected files and intended AI use. Available provenance and applicable releases are reviewed before approval; permitted uses are defined in the agreement.
Request a Human Activity Recognition Dataset
Send the target activities, model task, sequence requirements, participant and object criteria, environments, viewpoints, format, volume, annotations, split rules, rights requirements, and delivery format. Wavebreak Media will assess ready-made, archive-curated, and custom options against the specification.
