Nexar Dashcam Collision Prediction Dataset and Challenge

TL;DR

The paper introduces the Nexar Dashcam Collision Prediction Dataset and Challenge to advance traffic accident anticipation for autonomous driving. It provides 1,500 real-world dashcam clips with comprehensive temporal annotations, including the time of alert and time of accident, and defines a time-to-event based evaluation framework. The contributions include a robust, privacy-preserving dataset, clear acceptance criteria, and a public Kaggle benchmark designed to reward early and reliable predictions. This work offers a valuable, real-world resource and standardized testing protocol to push forward proactive collision prevention in ADAS and autonomous systems.

Abstract

This paper presents the Nexar Dashcam Collision Prediction Dataset and Challenge, designed to support research in traffic event analysis, collision prediction, and autonomous vehicle safety. The dataset consists of 1,500 annotated video clips, each approximately 40 seconds long, capturing a diverse range of real-world traffic scenarios. Videos are labeled with event type (collision/near-collision vs. normal driving), environmental conditions (lighting conditions and weather), and scene type (urban, rural, highway, etc.). For collision and near-collision cases, additional temporal labels are provided, including the precise moment of the event and the alert time, marking when the collision first becomes predictable. To advance research on accident prediction, we introduce the Nexar Dashcam Collision Prediction Challenge, a public competition on top of this dataset. Participants are tasked with developing machine learning models that predict the likelihood of an imminent collision, given an input video. Model performance is evaluated using the average precision (AP) computed across multiple intervals before the accident (i.e. 500 ms, 1000 ms, and 1500 ms prior to the event), emphasizing the importance of early and reliable predictions. The dataset is released under an open license with restrictions on unethical use, ensuring responsible research and innovation.

Figures (5)

• Figure 1: Nexar dataset samples. First column represents Before accident interval, columns 2-4 represent Alert interval, and fifth column represent After accident interval. Within the alert interval, column 2 represents the "time-of-alert" which is the earliest moment that the driver could intervene to prevent the accident. column 4 represents the "time-of-accident", and column 3 represents an intermediate frame between "time-of-alert" and "time-of-accident".
• Figure 2: Our dataset contains a considerable fraction of samples where the accident happens within just a couple of frames, a time interval that is far shorter than a human driver could react. The anticipation for such cases is applicable only for autonomous driving scenarios, indicating that our dataset opens up the potential for evaluating algorithms that serve beyond human drivers.
• Figure 3: Our dataset demonstrates significant diversity regarding road types, weather condition, lighting conditions, types of vehicles as well as blurring, lens flare or other artifacts caused by the camera capture process, serving as a more challenging testbed than representative existing datasets for early traffic anticipation (CTAD luo2023simulation, CCD bao2020uncertainty, ROL karim2023attention).
• Figure 4: Our dataset demonstrates diversity regarding video recording with respect to the weather, lighting condition as well as road types. One of the video was recorded indoor and does not have the weather label.
• Figure 5: Histogram of video duration (left), time of event (center), and alert-to-accident interval (right).