ADAS-TO-Sample: 10% Stratified Preview of Real-World Driver Takeover Events

A 10% stratified sample of the ADAS-TO dataset, containing approximately 1,570 real-world video clips of driver takeover events. The clips capture moments where control transitions from an engaged Advanced Driver Assistance System (ADAS) to the human driver. The sample represents 163 vehicle models from 23 manufacturers and was created by HenryYHW, with a last update recorded in March 2026.

Use Cases

• Benchmarking driver takeover prediction models based on real-world video sequences.
• Analyzing human response patterns to ADAS disengagement in scenarios like oncoming traffic or night driving.
• Training computer vision models to recognize takeover cues from in-cabin or road-facing video feeds.
• Studying the diversity of takeover events across a wide range of vehicle models and manufacturers.

Strengths

• Contains approximately 1,570 real-world takeover event clips, providing a substantial sample for evaluation.
• Represents a diverse set of 163 vehicle models from 23 different manufacturers.
• Clips are focused on the takeover moment, showing ±3 seconds around the event for context.

Limitations

• Description metadata is limited; actual data quality requires manual inspection after download.
• Column-level documentation is absent; field semantics must be inferred after download.
• This is a 10% sample; the full dataset is a separate resource.

Provenance

Source

HenryYHW

Collection Method

Likely collected from real-world driving scenarios involving ADAS engagement.

Freshness

Last updated 2026-03-06 17:57:30; freshness should be verified.