Multi-Occlusion Head Rotation Video Dataset

This dataset contains head rotation videos of individuals under different occlusion conditions — wearing hats, masks, sunglasses, or no occlusion — covering natural head movements from multiple viewing angles. It is designed for robust face recognition testing, head pose estimation, behavioral analysis, and multimodal perception tasks.

Subjects perform smooth and complete head-turning actions in diverse backgrounds under standard lighting conditions, ensuring balanced data distribution across all occlusion types. Each participant completes a full sequence under all four states, with videos recorded in 1080p at 30 FPS. The dataset supports a variety of AI model training and evaluation applications.

This dataset contains over 20,000 high-quality annotated video clips. The estimated price is approximately €8,000, and can be adjusted based on delivery format, licensing scope, and customization requirements.

Keywords: head occlusion, mask recognition, hat recognition, sunglasses recognition, head movement, face detection

Resolution: 1920×1080

Frame Rate (FPS): 30 FPS

Video Format / Codec: MP4 (H.264)

Color Space: RGB

Capture Devices: high-definition webcam or smartphone camera (fixed frontal view)

Collection Conditions: captured under standard lighting (indoor or daylight) while participants perform head rotation actions (up, down, left, right, front-facing). Each occlusion condition (no occlusion, hat, mask, sunglasses) is recorded separately, with balanced representation across categories and diverse background scenes.

Total Videos: approximately 5 videos per subject, totaling over 15,000 clips in the current dataset; assuming 10,000 subjects, the dataset scales to over 50,000 clips

Total Duration: about 1 minute per subject (≈50 hours existing data, equivalent to 200+ hours of synthetic expansion) Number of Subjects: currently 3,000+, scalable to 10,000+ through face-swapping augmentation

Data Volume: roughly 0.5 GB per person; 10,000+ subjects yield ~5 TB total

Unique Individuals / Scenes: (recommended: specify number of subjects and unique backgrounds)

Annotation Status: annotated (customizable)

Annotation Type: facial keypoint trajectory + head pose angle sequence + occlusion category classification

Label Categories / Distribution: Occlusion state (hat, mask, sunglasses, none) × Pose angle (left / right / up / down / front) × Gender / hairstyle (optional)

Sample Video Screenshot：