VZCrash: A Large-Scale IMU Dataset of Ego-Vehicle Crashes

2026-06-04

AI Summary

This study addresses the limitation in existing vehicle crash detection research caused by the scarcity of large-scale, real-world in-vehicle IMU data. To bridge this gap, we introduce VZCrash, the largest publicly available dataset to date, comprising over 31,000 real-world crash events and 158,000 negative samples collected from 73,010 commercial vehicles operating across the United States over multiple years. The dataset includes 100 Hz IMU measurements (acceleration and angular velocity) alongside 1 Hz GPS speed data. Leveraging VZCrash, we systematically evaluate threshold-based heuristics and various deep learning models, providing the first empirical evidence of the critical role data scale plays in detection performance. Our findings demonstrate that large-scale real-world data substantially enhances model accuracy and robustness in practical deployment, establishing a reliable benchmark for intelligent transportation safety systems.

Abstract

We introduce VZCrash, the largest publicly available dataset of real-world vehicle collision data featuring Inertial Measurement Unit (IMU) telemetry. The dataset contains more than 31,000 validated crashes and 158,000 negative samples, including hard cases and distractors. Each sample includes acceleration and angular velocity at 100 Hz, and GPS speed at 1 Hz. Events in VZCrash were captured by devices installed on a fleet of 73,010 commercial vehicles of different sizes driving in the United States over the span of several years. We also present an extensive experimental study enabled by the volume of the dataset. We first benchmark several different approaches, from a simple threshold-based heuristic to state-of-the-art deep learning models. Then, we present an experiment demonstrating the importance of scaling data to train high-quality crash detection models, and we show that scale is especially important when these models need to be deployed into a real-world environment.

Problem

• vehicle crash detection
• IMU dataset
• real-world collision data
• large-scale dataset
• ego-vehicle crashes

Innovation

• IMU dataset
• vehicle crash detection
• large-scale data
• deep learning benchmarking
• real-world deployment