Dataset Card for PointCloudPeople

The use of light detection and ranging (LiDAR) sensor technology for people detection offers a significant advantage in terms of data protection. However, to design these systems cost- and energy-efficiently, the relationship between the measurement data and final object detection output with deep neural networks (DNNs) has to be elaborated. Therefore, we present an automatically labeled LiDAR dataset for person detection, with different disturbance objects and various augmentation methods to analyze the influence of the LiDAR sensor parameters distance, resolution, noise, and shading in real point clouds.

Dataset Details

The data set is structured as follows. As described in the corresponding paper, an original data set consisting of point clouds in .pcd format was recorded with a Blickfeld Qb2 LiDAR sensor and automatically labeled in .json format using GDPR-compliant cameras (Basic). This dataset was then augmented with different sensor and mounting parameters (distance, noise, resolution, shading).

• Curated by: Lukas Haas
• License: CC-BY-4.0

Dataset Sources

The data set consists of approx—2000 point clouds with the corresponding labels per sub-data set collected with a Blickfeld Qb2 LiDAR Sensor. The augmentation generated a total of 95 derivatives of the dataset. The data set contains at least one person per point cloud and other interfering objects in the sensor's FOV. The subjects in the FoV are of different sizes.

Dataset Structure

Based on the measurement setup presented in the paper a dataset is recorded with the Qb2 to train the DNNs. Based on this dataset, different datasets with various resolutions, distances to the scene, noise, and shading of people in the point cloud of the LiDAR sensor are augmented to investigate the individual influence of the parameters on the object detection performance of DNNs. The point clouds are in .pcd files with the x, y, z, intensit format, and the labels in .json files in the unified 3D box definition label format for each object.

Personal and Sensitive Information

Due to the anonymity of people in LiDAR point clouds, personal data is protected. Furthermore, GDPR conform cameras are used to label the point clouds automatically.

Bias, Risks, and Limitations

Considering the limitations of our dataset, real-world tests should be conducted. GDPR conforms with care in a safe environment. However, it should be noted that although the data set includes people of different body sizes, the majority of people are adults due to the location of the measurement setup.

Citation

BibTeX:

@Article{s25103114, AUTHOR = {Haas, Lukas and Sanne, Florian and Zedelmeier, Johann and Das, Subir and Zeh, Thomas and Kuba, Matthias and Bindges, Florian and Jakobi, Martin and Koch, Alexander W.}, TITLE = {LiDAR Sensor Parameter Augmentation and Data-Driven Influence Analysis on Deep-Learning-Based People Detection}, JOURNAL = {Sensors}, VOLUME = {25}, YEAR = {2025}, NUMBER = {10}, ARTICLE-NUMBER = {3114}, URL = {https://www.mdpi.com/1424-8220/25/10/3114}, PubMedID = {40431906}, ISSN = {1424-8220}, DOI = {10.3390/s25103114} }

APA:

Haas, L., Sanne, F., Zedelmeier, J., Das, S., Zeh, T., Kuba, M., Bindges, F., Jakobi, M., & Koch, A. W. (2025). LiDAR Sensor Parameter Augmentation and Data-Driven Influence Analysis on Deep-Learning-Based People Detection. Sensors, 25(10), 3114. https://doi.org/10.3390/s25103114