A Small-Object-Detection Algorithm Based on LiDAR Point-Cloud Clustering for Autonomous Vehicles
Abstract
:1. Introduction
- A segmented ground-point cloud accurate segmentation algorithm is proposed. The object point clouds such as special points, approach points, object internal points, and departure points are filtered out according to the heuristic rules. According to the near-dense and far-sparse distributions of the point clouds, a multi-region polar coordinate meshing is adopted, and the ground segmentation is realized by plane-fitting. The proposed segmentation algorithm is qualitatively evaluated and compared with random sample consensus (RANSAC) and region-wise ground plane fitting (R-GPF). The proposed ground-segmentation algorithm can solve the problem of it being difficult to accurately segment the object point cloud and the ground-point cloud in different scenarios;
- In view of the shortcomings of the DBSCAN algorithm in terms of parameter sensitivity and poor real-time performance, an improved DBSCAN algorithm is proposed. The K-means++ algorithm is adopted for pre-clustering. The neighborhood radius is adaptively adjusted according to the distance, and the core point search method of the original algorithm is improved to reduce the number of traversal points. The improved DBSCAN algorithm can improve the clustering ability of adjacent small objects and long-distance small objects, and improve the clustering speed;
- The ground-segmentation algorithm is verified in a rough road-surface scene and a sloping road-surface scene, respectively. The clustering effect of the proposed improved DBSCAN algorithm on small-object point clouds is verified in real vehicle experiments. The experimental results showed that the improved small-object-detection algorithm based on the point cloud clustering method proposed in this paper can achieve the accurate detection of small objects.
2. Literature Review
2.1. Ground Segmentation
2.2. Point-Cloud Clustering
3. Methods
3.1. Ground-Point Cloud Segmentation
3.1.1. Coarse Filtering of Point Clouds
- (1)
- Special point filtering rules
- (2)
- Object point filtering rules
3.1.2. Multi-Area Polar Coordinate Meshing
3.1.3. Selection of Seed-Point Sets
3.1.4. Planar Model Fitting
3.2. Small Objective Clustering Based on an Improved DBSCAN Algorithm
- (1)
- Neighborhood: For any point , the circular region with as the center and the neighborhood radius (Eps) as the radius is called the neighborhood of ;
- (2)
- Density Threshold (MinPts): The minimum number of points required in a neighborhood for a point to be considered a core point is called the MinPts;
- (3)
- Core Point: For any point , if the number of points in its neighborhood exceeds the MinPts, then is a core point;
- (4)
- Border Point: For any point , if the number of points in its neighborhood is less than the MinPts, then is a border point;
- (5)
- Noise Point: If a point does not belong to any cluster, then is considered a noise point.
- (1)
- Before executing the DBSCAN algorithm, the k-means ++ algorithm is introduced to precluster the dataset. Any point from the point cloud data is randomly selected as the initial cluster center , and the probability that the remaining data points become the cluster center is calculated according to Equation (10).
- (2)
- Before performing refined clustering, the distribution characteristics of the point cloud that is dense nearby and sparse far away are used to select the appropriate neighborhood radius for data points with different distances. According to the relationship between the distance and neighborhood radius, the adaptive neighborhood radius is used, and its value can be calculated using Equation (11):
- (3)
- The main reason for the time-consuming nature of the traditional DBSCAN algorithm is that it must traverse all the data points; therefore, to improve the operational efficiency of the algorithm, we propose an improved method for searching for core points. First, any point in the point cloud data is selected as the starting point, the points exceeding the density threshold MinPts [18] are categorized as core points, and all the points in their neighborhood are removed. This step is repeated for the removed point cloud until all the core points are found. Finally, to avoid the existence of core points in the removed point cloud, the core points are considered as individual clusters, and intersecting clusters are merged. If there is a data point within the intersecting neighborhood space, the data point is categorized as a core point, and if there is a data point within the intersecting neighborhood space, the data point is categorized as a core point. If a data point exists within the intersecting neighborhood space, it is categorized as a core point and grouped into a cluster. As shown in Figure 7, because points A, B, D, and E satisfy the conditions of core points, they are judged as core points, but there point C is in the intersection region of the neighborhoods of points A and B. Point C is taken as a new core point, and because there is no data point in the intersection region of the neighborhoods of points D and E, the process of searching for core points is completed.
- (4)
- The points in the neighborhood space of the core points are grouped into the same cluster as the core points to complete the intragroup clustering. To avoid the problem of the same cluster being split into two clusters based on grouping, the intergroup distance is first checked. If it is smaller than the specified threshold, then it indicates that there may be a clustering split; if the distance between the clusters is smaller than the distance threshold, then it indicates that the two clusters are in the same category, and the two clusters are combined into the same cluster. If the intergroup distance is larger than the specified threshold, then it indicates that there is no correlation between the two groups, and no further follow-up check is performed.
3.3. Construction of 3D Bounding Boxes
4. Experiments and Results
4.1. Setup of Experimental Environment
4.2. Ground-Segmentation Experiment and Results Analysis
4.3. Cluster Detection Algorithm Experiments and Results Analysis
5. Discussion and Conclusions
Author Contributions
Funding
Institutional Review Board Statement
Informed Consent Statement
Data Availability Statement
Acknowledgments
Conflicts of Interest
References
- Ackerman, E. Lidar that will make self-driving cars affordable [News]. IEEE Spectrum 2016, 53, 14. [Google Scholar] [CrossRef]
- Shi, W.; Rajkumar, R. Point-gnn: Graph neural network for 3d object detection in a point cloud. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Seattle, WA, USA, 13–19 June 2020; pp. 1711–1719. [Google Scholar]
- Fernandes, D.; Silva, A.; Névoa, R.; Simões, C.; Gonzalez, D.; Guevara, M.; Novais, P.; Monteiro, J.; Melo-Pinto, P. Point-cloud based 3D object detection and classification methods for self-driving applications: A survey and taxonomy. Inf. Fusion. 2021, 68, 161–191. [Google Scholar] [CrossRef]
- Shi, S.; Wang, Z.; Shi, J.; Wang, X.; Li, H. From points to parts: 3d object detection from point cloud with part-aware and part-aggregation network. IEEE Trans. Pattern Anal. Mach. Intell. 2020, 43, 2647–2664. [Google Scholar] [CrossRef]
- Qi, C.R.; Su, H.; Mo, K.; Guibas, L.J. Pointnet: Deep learning on point sets for 3d classification and segmentation. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017; pp. 652–660. [Google Scholar]
- Zhou, Y.; Tuzel, O. Voxelnet: End-to-end learning for point cloud based 3d object detection. In Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), Salt Lake City, UT, USA, 18–22 June 2018; pp. 4490–4499. [Google Scholar]
- Lang, A.H.; Vora, S.; Caesar, H.; Zhou, L.; Yang, J.; Beijbom, O. Pointpillars: Fast encoders for object detection from point clouds. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA, 15–20 June 2019; pp. 12697–12705. [Google Scholar]
- Yan, Y.; Mao, Y.; Li, B. Second: Sparsely embedded convolutional detection. Sensors 2018, 18, 3337. [Google Scholar] [CrossRef] [PubMed]
- Shi, S.; Wang, X.; Li, H. Pointrcnn: 3d object proposal generation and detection from point cloud. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, Long Beach, CA, USA, 15–20 June 2019; pp. 770–779. [Google Scholar]
- Redmon, J.; Divvala, S.; Girshick, R.; Farhadi, A. You only look once: Unified, real-time object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Las Vegas, NV, USA, 27–30 June 2016; IEEE: Piscataway, NJ, USA, 2016; pp. 779–788. [Google Scholar]
- Gao, H.; Cheng, B.; Wang, J.; Li, K.; Zhao, J.; Li, D. Object classification using CNN-based fusion of vision and LIDAR in autonomous vehicle environment. IEEE Trans. Ind. Inform. 2018, 14, 4224–4231. [Google Scholar] [CrossRef]
- Lin, T.Y.; Dollár, P.; Girshick, R.; He, K.; Hariharan, B.; Belongie, S. Feature pyramid networks for object detection. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition, Honolulu, HI, USA, 21–26 July 2017; pp. 2117–2125. [Google Scholar]
- Ren, S.; He, K.; Girshick, R.; Sun, J. Faster r-cnn: Towards real-time object detection with region proposal networks. In Proceedings of the Advances in Neural Information Processing Systems 28 (NIPS 2015), Montreal, QC, Canada, 7–12 December 2015; Volume 28, pp. 91–99. [Google Scholar]
- Himmelsbach, M.; Hundelshausen, F.V.; Wuensche, H.J. Fast segmentation of 3D point clouds for ground vehicles. In Proceedings of the 2010 IEEE Intelligent Vehicles Symposium, La Jolla, CA, USA, 21–24 June 2010; pp. 560–565. [Google Scholar]
- Weinmann, M.; Jutzi, B.; Mallet, C. Semantic 3D scene interpretation: A framework combining optimal neighborhood size selection with relevant features. In Proceedings of the ISPRS Annals of Photogrammetry, Remote Sensing and Spatial Information Sciences, Zurich, Switzerland, 5–7 September 2014; pp. 181–188. [Google Scholar]
- Rusu, R.B.; Blodow, N.; Beetz, M. Fast point feature histograms (FPFH) for 3D registration. In Proceedings of the 2009 IEEE International Conference on Robotics and Automation, Kobe, Japan, 12–17 May 2009; pp. 3212–3217. [Google Scholar]
- Hu, Y. Research on Three-Dimensional Small Object Detection Method Based on Attention Feature Enhancement. Master’s Thesis, Jianghan University, Wuhan, China, 2023. [Google Scholar]
- Ester, M.; Kriegel, H.P.; Sander, J.; Xu, X. Density-based spatial clustering of applications with noise. In Proceedings of the International Conference on Knowledge Discovery and Data Mining, Portland, OR, USA, 2–4 August 1996; p. 240. [Google Scholar]
- Arthur, D.; Vassilvitskii, S. k-Means++: The Advantages of Careful Seeding; Stanford InfoLab Publication Server: Online, 2007; Volume 7, pp. 1027–1035. [Google Scholar]
- Qu, W. Research on Obstacle Detection Technology of Engineering Vehicles in Unstructured Environment. Master’s Thesis, Jilin University, Changchun, China, 2023. (In Chinese). [Google Scholar] [CrossRef]
- Arora, M.; Wiesmann, L.; Chen, X.; Stachniss, C. Static map generation from 3D LiDAR point clouds exploiting ground segmentation. Robot. Auton. Syst. 2023, 159, 104287. [Google Scholar] [CrossRef]
- Cheng, Z.; Ren, G.; Zhang, Y. Ground segmentation algorithm based on 3D LiDAR point cloud. In Proceedings of the 2018 International Conference on Mechanical, Electrical, Electronic Engineering & Science (MEEES 2018), Chongqing, China, 26–27 May 2018; Atlantis Press: Amsterdam, The Netherlands, 2018; pp. 16–21. [Google Scholar]
- Zuo, Z.; Fu, Z.; Li, Z.; Wang, Y.; Ren, Y. Ground Segmentation of 3D LiDAR Point Cloud with Adaptive Threshold. In Proceedings of the 2023 42nd Chinese Control Conference (CCC), Tianjin, China, 24–26 July 2023; IEEE: Piscataway, NJ, USA, 2023; pp. 8032–8037. [Google Scholar]
- Yan, Y.; Chen, Z.; Sun, Y.; Li, Z.; Yao, C. Ground filtering algorithm of LiDAR point cloud in dense low vegetation area. Bull. Surv. Mapp. 2021, 07, 1–5. (In Chinese) [Google Scholar] [CrossRef]
- Lim, H.; Hwang, S.; Myung, H. ERASOR: Egocentric ratio of pseudo occupancy-based dynamic object removal for static 3D point cloud map building. IEEE Robot. Autom. Lett. 2021, 6, 2272–2279. [Google Scholar] [CrossRef]
- Li, E.; Razani, R.; Xu, Y.; Liu, B. Cpseg: Cluster-free panoptic segmentation of 3d lidar point clouds. In Proceedings of the 2023 IEEE International Conference on Robotics and Automation (ICRA), London, UK, 29 May–2 June 2023; IEEE: Piscataway, NJ, USA, 2023; pp. 8239–8245. [Google Scholar]
- Fang, B.; Zhang, X.; Wang, H. Pixel-level segmentation algorithm combining depth map clustering and object detection. J. Pattern Recognit. Artif. Intell. 2022, 35, 130–140. (In Chinese) [Google Scholar] [CrossRef]
- Cai, J. Vehicle Detection Based on LiDAR Point Cloud Data. Master’s Thesis, Harbin Institute of Technology, Harbin, China, 2020. [Google Scholar] [CrossRef]
- Sun, B.; Li, W.; Liu, H.; Wang, P.; Gao, S.; Feng, P. Mathematical method for Lidar-based obstacle detection of intelligent vehicle. IAENG Int. J. Comput. Sci. 2021, 48, 181–189. [Google Scholar]
- Zhou, W.; Wang, L.; Han, X.; Zhang, X.; Chen, Y. Adaptive density spatial clustering method fusing chameleon swarm algorithm. Entropy 2023, 25, 782. [Google Scholar] [CrossRef] [PubMed]
- Falahiazar, Z.; Bagheri, A.; Reshadi, M. Determining the parameters of DBSCAN automatically using the multi-objective genetic algorithm. J. Inf. Sci. Eng. 2021, 37, 157–183. [Google Scholar] [CrossRef]
- Sun, Z.; Li, Z.; Liu, Y. An improved Lidar data segmentation algorithm based on Euclidean clustering. In Proceedings of the 11th International Conference on Modelling, Identification and Control (ICMIC2019); Springer: Singapore, 2020; pp. 1119–1130. [Google Scholar]
- Qu, J.; Li, S.; Li, Y.; Liu, L. Research on railway obstacle detection method based on developed Euclidean clustering. Electronics 2023, 12, 1175. [Google Scholar] [CrossRef]
- Gottschalk, S.; Lin, M.C.; Manocha, D. OBBTree: A hierarchical structure for rapid interference detection. In Proceedings of the 23rd Annual Conference on Computer Graphics and Interactive Techniques; Association for Computing Machinery: New York, NY, USA, 1996; pp. 171–180. [Google Scholar]
- Mei, S.M.; Huang, M.H.; Liu, Z.H. Ground segmentation method in complex scenes based on three-dimensional LiDAR. Laser Optoelectron. Prog. 2022, 59, 422–429. [Google Scholar]
- Fischler, M.A.; Bolles, R.C. Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography. Commun. ACM 1981, 24, 381–395. [Google Scholar] [CrossRef]
Parameters | Values |
---|---|
Laser Harness (Wire) | 32 |
Measuring range (m) | 80–100 |
Accuracy (cm) | ±2 |
Field of view (°) | 360 41.34 |
Supply Voltage (VDC) | 9–32 |
Power (W) | 31.4 |
Output (points per second) | 700,000 |
Parameter | Value | Parameter | Value |
---|---|---|---|
20 | 3.3 | ||
a | 0.025 | 0.2 m | |
b | −0.05 | 5 |
Algorithms | Accuracy Rate A | F1 Score F1 | Average Running Time | FPS | ||
---|---|---|---|---|---|---|
RANSAC | 91.66% | 86.52% | 85.82% | 89.02% | 69.52 ms | 14.4 |
R-GPF | 89.26% | 91.33% | 89.27% | 90.29% | 28.55 ms | 35.0 |
Ours | 91.86% | 92.70% | 90.94% | 92.28% | 32.47 ms | 30.8 |
Clustering Algorithm | Object Number | Number of Correct Detections | Recall Rate | Average Running Time | FPS |
---|---|---|---|---|---|
DBSCAN | 985 | 739 | 75.03% | 57.85 ms | 17.3 |
Ours | 985 | 832 | 94.47% | 49.89 ms | 20.0 |
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content. |
© 2024 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https://creativecommons.org/licenses/by/4.0/).
Share and Cite
Duan, Z.; Shao, J.; Zhang, M.; Zhang, J.; Zhai, Z. A Small-Object-Detection Algorithm Based on LiDAR Point-Cloud Clustering for Autonomous Vehicles. Sensors 2024, 24, 5423. https://doi.org/10.3390/s24165423
Duan Z, Shao J, Zhang M, Zhang J, Zhai Z. A Small-Object-Detection Algorithm Based on LiDAR Point-Cloud Clustering for Autonomous Vehicles. Sensors. 2024; 24(16):5423. https://doi.org/10.3390/s24165423
Chicago/Turabian StyleDuan, Zhibing, Jinju Shao, Meng Zhang, Jinlei Zhang, and Zhipeng Zhai. 2024. "A Small-Object-Detection Algorithm Based on LiDAR Point-Cloud Clustering for Autonomous Vehicles" Sensors 24, no. 16: 5423. https://doi.org/10.3390/s24165423