Nexar Collision Prediction: 1,500 Dashcam Videos for Crash Detection

1,500 dashcam videos from Nexar, with half showing collisions or near-collision events. Videos have a resolution of 1280x720 at 30 frames per second and are typically about 40 seconds long. This dataset was created by nexar-ai for a Kaggle challenge and was last updated on June 10, 2025.

Use Cases

• Training collision prediction models based on dashcam video sequences.
• Developing anomaly detection for road safety based on imminent collision events.
• Benchmarking video classification algorithms on a balanced set of positive and negative driving scenarios.
• Analyzing driver behavior and pre-crash events from real-world dashcam footage.

Strengths

• Balanced dataset with 750 positive (collision/near-collision) and 750 negative (regular driving) videos.
• Consistent video format: 1280x720 resolution at 30 frames per second.
• Real-world data sourced from Nexar dashcams in vehicles.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Freshness should be verified as the last update was 2025-06-10.

Provenance

Source

Nexar dashcams, collected by nexar-ai.

Collection Method

Videos recorded from dashcams installed in vehicles.

Time Range

null

Freshness

Last updated 2025-06-10 22:09:08.

Geography

null