Bosch Street Dataset: A Multi-Modal Dataset with Imaging Radar for Automated Driving

TL;DR

The Bosch Street Dataset (BSD) addresses the lack of large-scale, high-resolution imaging radar data for HAD and ADAS by providing a multi-modal sensor belt (9 imaging radars, 4 lidars, 4 cameras) with full 360° coverage across urban, rural, and highway scenes. It combines extensive manual and auto-labeling, a dedicated development kit, and baseline benchmarks to enable robust radar-centric perception and sensor fusion research. Key findings show that imaging radar density and dataset scale substantially close the gap between radar and lidar performance, and that auto-labeling can scale annotations without sacrificing quality. BSD thus enables more reliable radar-based perception, supports cross-modal fusion studies, and is positioned to catalyze collaboration between Bosch and the HAD/ADAS research community.

Abstract

This paper introduces the Bosch street dataset (BSD), a novel multi-modal large-scale dataset aimed at promoting highly automated driving (HAD) and advanced driver-assistance systems (ADAS) research. Unlike existing datasets, BSD offers a unique integration of high-resolution imaging radar, lidar, and camera sensors, providing unprecedented 360-degree coverage to bridge the current gap in high-resolution radar data availability. Spanning urban, rural, and highway environments, BSD enables detailed exploration into radar-based object detection and sensor fusion techniques. The dataset is aimed at facilitating academic and research collaborations between Bosch and current and future partners. This aims to foster joint efforts in developing cutting-edge HAD and ADAS technologies. The paper describes the dataset's key attributes, including its scalability, radar resolution, and labeling methodology. Key offerings also include initial benchmarks for sensor modalities and a development kit tailored for extensive data analysis and performance evaluation, underscoring our commitment to contributing valuable resources to the HAD and ADAS research community.

Figures (8)

• Figure 1: The Bosch street dataset (BSD) showcased with sample data captures. The setup includes 9 imaging radar sensors, 4 cameras, and a 64-channel lidar sensor, covering a complete 360-degree field of view. Data from varied settings —urban, rural, and highways — highlight the dataset's versatility across different geographical locations and under various weather and lighting conditions. The dataset features 13.6k sequences and over 36 hours of measurements, all annotated with bounding box labels to facilitate advanced HAD research.
• Figure 2: Bird's eye view of the sensor setup. (a) shows the cameras and (b) shows the lidars and the radars. The sensors are depicted by means of of their local coordinate frames. Axes according to ISO 8855. Vehicle not to scale.
• Figure 3: Example scenes from the dataset. The first column shows the camera image, the middle column lidar, and the right column radar. The rows represent various environmental conditions from top to bottom: a nighttime scene, a rainy scene, an intersection, a tunnel, and a roundabout.
• Figure 4: Distribution of the train, validation, and test split of the dataset. The dataset is diverse with respect to location, weather, time of day, and driving situation. BW stands for Baden-Württemberg and SF stands for San Francisco Bay area. The test dataset contains more manually annotated data, to ensure highest data quality.
• Figure 5: Distribution of labels (for the classes also shown in Fig. \ref{['fig:data_piechart']}) over distance (range) on the left and velocity on the right comparing the BSD dataset to nuScenes. Density on the y-axis was computed as histograms, normalized to have an area of 1. This allows to compare coverage and shows that our BSD also provides instances for distant labels, i.e. beyond 100 m, and fast labels, i.e. beyond 20 m/s.