Zitao Lin

Zitao Lin


Research Focus & Metrics

Academic Career Length

4Years (2023-Present)

(Calculated once Scilit publication data is synced: current year − year of first publication + 1)

Representative Publications

View All Publications
4 June 2023 in Remote Sensing
Verified
Journal Article
Open access
PDF file
CKChuanli Kang
Chuanli Kang
ZLZitao Lin
Zitao Lin
SWSiyi Wu
Siyi Wu
YLYiling Lan
Yiling Lan
崇耿崇铭 耿
崇铭 耿
SZSai Zhang
Sai Zhang
High-precision ground point cloud data has a wide range of applications in various fields, and the separation of ground points from non-ground points is a crucial preprocessing step. Therefore, designing an efficient, accurate, and stable ground extraction algorithm is highly significant for improving the processing efficiency and analysis accuracy of point cloud data. The study area in this article was a park in Guilin, Guangxi, China. The point cloud was obtained by utilizing the UAV platform. In order to improve the stability and accuracy of the filter algorithm, this article proposed a triangular grid filter based on the Slope Filter, found violation points by the spatial position relationship within each point in the triangulation network, improved KD-Tree-Based Euclidean Clustering, and applied it to the non-ground point extraction. This method is accurate, stable, and achieves the separation of ground points from non-ground points. Firstly, the Slope Filter is used to remove some non-ground points and reduce the error of taking ground points as non-ground points. Secondly, a triangular grid based on the triangular relationship between each point is established, and the violation triangle is determined through the grid; thus, the corresponding violation points are found in the violation triangle. Thirdly, according to the three-point collinear method to extract the regular points, these points are used to extract the regular landmarks by the KD-Tree-Based Euclidean Clustering and Convex Hull Algorithm. Finally, the dispersed points and irregular landmarks are removed by the Clustering Algorithm. In order to confirm the superiority of this algorithm, this article compared the filter effects of various algorithms on the study area and filtered the 15 data samples provided by ISPRS, obtaining an average error of 3.46%. The results show that the algorithm presented in this article has high processing efficiency and accuracy, which can significantly improve the processing efficiency of point cloud data in practical applications.

14 March 2024 in Sensors
Verified
Journal Article
Open access
PDF file
CKChuanli Kang
Chuanli Kang
崇耿崇铭 耿
崇铭 耿
ZLZitao Lin
Zitao Lin
SZSai Zhang
Sai Zhang
SZSiyao Zhang
Siyao Zhang
SWShiwei Wang
Shiwei Wang
Existing point-to-point registration methods often suffer from inaccuracies caused by erroneous matches and noisy correspondences, leading to significant decreases in registration accuracy and efficiency. To address these challenges, this paper presents a new coarse registration method based on a geometric constraint and a matrix evaluation. Compared to traditional registration methods that require a minimum of three correspondences to complete the registration, the proposed method only requires two correspondences to generate a transformation matrix. Additionally, by using geometric constraints to select out high-quality correspondences and evaluating the matrix, we greatly increase the likelihood of finding the optimal result. In the proposed method, we first employ a combination of descriptors and keypoint detection techniques to generate initial correspondences. Next, we utilize the nearest neighbor similarity ratio (NNSR) to select high-quality correspondences. Subsequently, we evaluate the quality of these correspondences using rigidity constraints and salient points’ distance constraints, favoring higher-scoring correspondences. For each selected correspondence pair, we compute the rotation and translation matrix based on their centroids and local reference frames. With the transformation matrices of the source and target point clouds known, we deduce the transformation matrix of the source point cloud in reverse. To identify the best-transformed point cloud, we propose an evaluation method based on the overlap ratio and inliers points. Through parameter experiments, we investigate the performance of the proposed method under various parameter settings. By conducting comparative experiments, we verified that the proposed method’s geometric constraints, evaluation methods, and transformation matrix computation consistently outperformed other methods in terms of root mean square error (RMSE) values. Additionally, we validated that our chosen combination for generating initial correspondences outperforms other descriptor and keypoint detection combinations in terms of the registration result accuracy. Furthermore, we compared our method with several feature-matching registration methods, and the results demonstrate the superior accuracy of our approach. Ultimately, by testing the proposed method on various types of point cloud datasets, we convincingly established its effectiveness. Based on the evaluation and selection of correspondences and the registration result’s quality, our proposed method offers a solution with fewer iterations and higher accuracy.

28 January 2024 in Sensors
Verified
Journal Article
Open access
PDF file
ZLZitao Lin
Zitao Lin
CKChuanli Kang
Chuanli Kang
SWSiyi Wu
Siyi Wu
XLXuanhao Li
Xuanhao Li
LCLei Cai
Lei Cai
SWShiwei Wang
Shiwei Wang
The point cloud segmentation method plays an important role in practical applications, such as remote sensing, mobile robots, and 3D modeling. However, there are still some limitations to the current point cloud data segmentation method when applied to large-scale scenes. Therefore, this paper proposes an adaptive clustering segmentation method. In this method, the threshold for clustering points within the point cloud is calculated using the characteristic parameters of adjacent points. After completing the preliminary segmentation of the point cloud, the segmentation results are further refined according to the standard deviation of the cluster points. Then, the cluster points whose number does not meet the conditions are further segmented, and, finally, scene point cloud data segmentation is realized. To test the superiority of this method, this study was based on point cloud data from a park in Guilin, Guangxi, China. The experimental results showed that this method is more practical and efficient than other methods, and it can effectively segment all ground objects and ground point cloud data in a scene. Compared with other segmentation methods that are easily affected by parameters, this method has strong robustness. In order to verify the universality of the method proposed in this paper, we test a public data set provided by ISPRS. The method achieves good segmentation results for multiple sample data, and it can distinguish noise points in a scene.

30 June 2023 in Sensors
Verified
Journal Article
Open access
PDF file
CKChuanli Kang
Chuanli Kang
ZLZitao Lin
Zitao Lin
SWSiyi Wu
Siyi Wu
JYJiale Yang
Jiale Yang
SZSiyao Zhang
Siyao Zhang
XLXuanhao Li
Xuanhao Li
With the rapid development of Lidar technology, the use of Lidar for underwater terrain detection has become feasible. There is still a challenge in the process of signal resolution: the underwater laser echo signal is different to propagating in the air, and it is easy to produce weak waves and superimposed waves. However, existing waveform decomposition methods are not effective in processing these waveform signals, and the underwater waveform signal cannot be correctly decomposed, resulting in subsequent data-processing errors. To address these issues, this study used a drone equipped with a 532 nm laser to detect a pond as the study background. This paper proposes an improved inflection point selection decomposition method to estimate the parameter. By comparing it with other decomposition methods, we found that the RMSE is 2.544 and R2 is 0.995975, which is more stable and accurate. After estimating the parameters, this study used oscillating particle swarm optimization (OPSO) and the Levenberg–Marquardt algorithm (LM) to optimize the estimated parameters; the final results show that the method in this paper is closer to the original waveform. In order to verify the processing effect of the method on complex waveform, this paper decomposes and optimizes the simulated complex waveforms; the final RMSE is 0.0016, R2 is 1, and the Gaussian component after decomposition can fully represent the original waveform. This method is better than other decomposition methods in complex waveform decomposition, especially regarding weak waves and superimposed waves.

11 February 2026 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
JDJiaqi Dong
Jiaqi Dong
GCGuoliang Chen
Guoliang Chen
MPMarco Piras
Marco Piras
YBYuewei Bo
Yuewei Bo
MHMinghong Hu
Minghong Hu
HLHu Liu
Hu Liu
Self-supervised monocular depth estimation leverages photometric consistency across views as supervisory signals, yet inherently suffers from compromised geometric fidelity at object boundaries, surface smoothness, and cross-view consistency due to 3-D structural information loss during 2-D projection. This article introduces Diff3D-Net, a novel framework integrating multilevel 3-D geometric constraints to overcome these limitations. Key innovations include: a fixed-parameter RPM-Net point cloud registration module generating geometrically grounded supervision; a pose consistency constraint that directly optimizes the pose network using RPM-Net-refined transformations to suppress error accumulation; a dual-path photometric constraint adaptively fusing reconstructions from original and optimized poses to enhance low-texture robustness; and a multiview 3-D structural consistency constraint enforcing global point alignment via minimum residual selection. Evaluations on KITTI and DDAD show Diff3D-Net’s state-of-the-art performance, achieving 0.094/4.220 Abs. Rel/root mean squared error (RMSE) on KITTI. The model also exhibits strong cross-domain generalization in zero-shot settings, outperforming competitors on the Cityscapes, NYU Depth V2, and DrivingStereo datasets without fine-tuning, highlighting its Internet of Thing (IoT) application potential. Despite higher training costs, its inference achieves an optimal speed–accuracy tradeoff, with a lightweight variant reaching 51.73 FPS—sufficient to enable real-time applications such as UAV navigation on resource-limited IoT edge devices. Ablations confirm each constraint’s contribution, and qualitative results demonstrate robustness in dynamic scenes and complex geometries.

27 May 2026 in IEEE Internet of Things Journal
Verified
Journal Article
Closed access
TWTeng Wang
Teng Wang
GCGuoliang Chen
Guoliang Chen
JBJingxue Bi
Jingxue Bi
JDJiaqi Dong
Jiaqi Dong
ZLZitao Lin
Zitao Lin
JLJie Liu
Jie Liu
With the evolution of Fifth-Generation (5G) wireless systems, high-resolution channel observations enabled by large bandwidths and massive Multiple-Input Multiple-Output (MIMO) have opened new opportunities for indoor positioning. However, while existing fingerprinting-based learning methods excel in complex multipath environments, they often rely on fixed anchor configurations. This makes them vulnerable to dynamic anchor visibility caused by signal blockages, user mobility, or hardware failures in practical deployments. To address this, we propose 5G Graph-based Localization (5GLoc), a learning-driven framework designed for robust indoor positioning under dynamic anchor availability. By modeling localization instances as graphs, 5GLoc utilizes a Dynamic Graph Attention Network (DGAT) to flexibly accommodate varying numbers and combinations of anchors, ensuring robust inference under heterogeneous observations. We further introduce a Combinatorial Dynamic Node Masking (CDNM) strategy for data augmentation, as well as a composite loss function with learnable weights for balancing accuracy and stability. Finally, a map-constrained Particle Filter (PF) is integrated to refine trajectories for spatial feasibility and trajectory consistency. Experimental results based on the IPIN 2025 dataset demonstrate that 5GLoc achieves superior performance in complex warehouse environments, with a 75th percentile error of 0.857 m. Compared to several representative positioning methods, 5GLoc reduces the Mean Absolute Error (MAE) and Root Mean Square Error (RMSE) by 42.2%–80.2% and 50.6%–80.8%, respectively, maintaining high reliability even under extreme conditions with reduced anchor availability. This framework provides an effective solution for 5G localization in dynamic environments, meeting both precision and real-time requirements.