MEVA: Multiview Human Activity Video with 328 Hours of Footage

Multiview video data of human activity, both scripted and unscripted, collected with roughly 100 actors over several weeks. The current release consists of about 328 hours (516GB, 4259 clips) of video data from 29 cameras, plus 4.6 hours of UAV data and annotations for roughly 184 hours. The dataset was created by Kitware, with further updates planned.

Use Cases

• Training multiview human activity recognition models based on scripted and unscripted video data.
• Developing camera calibration and sensor fusion techniques based on overlapping and non-overlapping camera fields of view.
• Researching person re-identification and tracking across multiple camera views.
• Benchmarking UAV-based activity detection algorithms based on the included aerial video data.

Strengths

• Contains about 328 hours of video data across 4259 clips, providing substantial volume for training.
• Includes annotations for roughly 184 hours of the video data, supporting supervised learning tasks.
• Captures multiview perspectives from 29 cameras with overlapping and non-overlapping fields of view.
• Incorporates multiple data modalities, including ground video, UAV footage, GPS tracks, and a site map.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• Data may reflect geographic or temporal bias inherent to the collection site and period.

Provenance

Source

Kitware

Collection Method

Video data collected with roughly 100 actors over several weeks using 29 cameras and UAVs.