RadarScenes: A Real-World Radar Point Cloud Data Set for Automotive Applications

Ole Schumann,Markus Hahn,Nicolas Scheiner,Fabio Weishaupt,Julius F. Tilly,Jürgen Dickmann,Christian Wöhler

2024-02-18

Abstract:A new automotive radar data set with measurements and point-wise annotations from more than four hours of driving is presented. Data provided by four series radar sensors mounted on one test vehicle were recorded and the individual detections of dynamic objects were manually grouped to clusters and labeled afterwards. The purpose of this data set is to enable the development of novel (machine learning-based) radar perception algorithms with the focus on moving road users. Images of the recorded sequences were captured using a documentary camera. For the evaluation of future object detection and classification algorithms, proposals for score calculation are made so that researchers can evaluate their algorithms on a common basis. Additional information as well as download instructions can be found on the website of the data set: <a class="link-external link-http" href="http://www.radar-scenes.com" rel="external noopener nofollow">this http URL</a>.

Machine Learning

What problem does this paper attempt to address?

### What problem does this paper attempt to solve?

This paper aims to solve the problem of the lack of large - scale, high - quality labeled datasets in the field of radar perception for autonomous vehicles. Specifically, the authors introduce a new dataset named **RadarScenes**, which contains more than 4 hours of driving records, covering 100 kilometers of diverse road scenes and having more than 7,500 uniquely manually - labeled road users. These labels involve 11 different object categories, including cars, large vehicles, trucks, buses, trains, bicycles, motorcycles, pedestrians, groups of pedestrians, animals and others.

#### Main problems and background

1. **Environmental perception in autonomous driving**:
- Autonomous vehicles need to perform high - precision and reliable perception of the surrounding environment. To achieve this, multiple sensors (such as cameras, lidars and radars) are usually combined to complement each other's advantages and disadvantages.
- Radars play an important role in autonomous driving systems because they work in the millimeter - wave band and are resistant to bad weather and can directly measure relative velocities (through the Doppler effect).

2. **Deficiencies of existing datasets**:
- Although significant progress has been made in the field of computer vision, especially in image processing, semantic scene - understanding algorithms based on radar data are still in their infancy.
- Currently, publicly available automotive radar datasets are either small in scale, lack diversity or use radar sensors that are not suitable for practical applications. This makes it difficult for researchers to compare the effects of different methods and limits the development of radar - perception algorithms.

3. **Promoting the development of radar - perception algorithms**:
- The introduction of the RadarScenes dataset aims to provide a standard platform for developing and evaluating new machine - learning (ML) radar - perception algorithms. In particular, for the tasks of detecting and classifying moving road users, this dataset provides rich point - cloud data and detailed labeling information.

#### Solutions

- **Large - scale, high - quality dataset**: RadarScenes contains data collected by four series of production radar sensors, covering multiple driving scenarios such as urban and rural areas, and labels 11 types of objects.
- **Diverse application scenarios**: This dataset is not only suitable for single - radar - perception tasks, but can also be used for pre - training multi - sensor - fusion networks.
- **Evaluation metrics**: The paper proposes several evaluation metrics (such as mAP, F1 - score, etc.) so that researchers can compare the performance of different algorithms on a common basis.

By providing such a comprehensive and diverse dataset, RadarScenes is expected to accelerate the research and development of radar - perception algorithms in the field of autonomous driving.