Multicamera synthetic dataset of crashes at the intersection

This dataset is a synthetic image dataset generated using Unreal Engine 5 and is designed to simulate traffic accidents at a road intersection, with a focus on car–car and car–pedestrian collisions. The dataset is intended for research on deep learning–based methods for traffic accident detection and prediction, particularly in the context of traffic surveillance applications. The dataset is based on a custom-built 3D intersection model inspired by a real-world intersection located in the Czech Republic. It contains four distinct scenes, where each scene represents a different part or viewpoint of the intersection. The scenes were created specifically for this dataset to provide realistic yet fully controlled traffic scenarios. For each scene, the dataset includes sequences depicting normal traffic behavior without any collisions, as well as sequences containing collision events. These collision scenarios include both vehicle–vehicle and vehicle–pedestrian accidents. In total, the dataset contains 2,192 images, of which 1,063 images represent normal traffic conditions and 1,129 images represent anomalous traffic with collision events. All images in the dataset are annotated with ground truth bounding boxes and corresponding class labels, enabling their direct use for supervised learning tasks such as object detection, tracking, and accident analysis. The dataset is provided to foster research and development of methods for automatic traffic collision detection, accident prediction, and anomaly detection in traffic surveillance systems. Warning: The dataset contains synthetic images depicting traffic collisions involving vehicles and pedestrians. Although the data is artificially generated, some images may be disturbing or uncomfortable for certain viewers. This dataset has been created as part of the SENDER project by Faculty of Electrical Engineering and Communication at Brno University of Technology (BUT) and Yunex s.r.o (part of Yunex Traffic) funded by Technology Agency of Czech Republic (TAČR).