PersonPath22: Large-Scale Multi-Person Tracking Dataset with 236 Videos

PersonPath22 is a large-scale multi-person tracking dataset containing 236 videos captured mostly from static-mounted cameras. The dataset was collected from sources where redistribution rights and participant consent were secured, and it is hosted by Amazon Web Services. Each video includes ground-truth annotations with bounding boxes and tracklet IDs for all persons in every frame.

Use Cases

• Train and evaluate multi-person tracking algorithms based on the provided bounding boxes and tracklet IDs.
• Benchmark pedestrian detection performance using the frame-level ground-truth annotations.
• Develop models for understanding crowd dynamics and movement patterns from static camera footage.

Strengths

• Contains 236 annotated videos, providing substantial volume for model training.
• Includes explicit consent and redistribution rights, addressing ethical and legal concerns.
• Provides frame-level ground-truth annotations with both bounding boxes and tracklet IDs for all persons.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count and video duration are unknown, which may limit suitability assessment.
• Last update date is unknown; freshness unverified.

Provenance

Source

Amazon Web Services

Collection Method

Videos were collected from sources where redistribution rights and participant consent were secured.

Time Range

null

Freshness

Last updated date is unknown.

Geography

null