Car Accident Video Dataset with 5,000 Crash Recordings

5,000 videos capturing traffic accident crashes, designed for detection and prediction tasks. The dataset was created by UniDataPro and last updated on Hugging Face in March 2026. It is engineered to challenge detection models and support traffic safety research.

Use Cases

• Training object detection models based on video footage of vehicle crashes.
• Developing traffic accident prediction systems based on crash event sequences.
• Benchmarking and challenging video-based detection models as described.
• Advancing traffic safety research based on structured crash data.

Strengths

• Contains 5,000 videos, providing a substantial collection for model training.
• Specifically engineered to challenge detection models, suggesting a focus on difficult cases.
• Last updated on 2026-03-30, indicating recent maintenance.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

UniDataPro on Hugging Face

Collection Method

Collection method is not specified in the provided description.

Time Range

Temporal coverage is not specified.

Freshness

2026-03-30

Geography

Spatial coverage is not specified, though one platform tag suggests 'Regionus'.