Online vehicle trajectory extraction based on LiDAR data
| dc.contributor.committeeChair | Liu, Hongchao | |
| dc.contributor.committeeMember | Fedler, Clifford | |
| dc.contributor.committeeMember | Xu, Hao | |
| dc.contributor.committeeMember | Won, Moon | |
| dc.creator | Zhang, Yibin | |
| dc.date.accessioned | 2024-01-04T20:56:10Z | |
| dc.date.available | 2024-01-04T20:56:10Z | |
| dc.date.issued | 2023-12 | |
| dc.description.abstract | This comprehensive research explores three distinct approaches for enhancing LiDAR data analysis and vehicle tracking in traffic monitoring and safety assessment applications. The first study introduces a novel background filtering technique, effectively identifying stationary vehicles amidst moving ones and significantly reducing the dataset size to improve computational efficiency. The second study proposes an unsupervised clustering method tailored for roadside LiDAR applications, utilizing a unique approach of converting 3D LiDAR data into a 2D data structure with the aid of LiDAR working principles This innovative approach, comprising a combination of region growing, connected component labeling, and an improved merge process, has proven effective in addressing over-segmentation challenges. The third section discusses the implementation and evaluation of a proposed index method for online vehicle tracking, outperforming the traditional bounding box method in accuracy and computation time. Inspired by the idea of integrating the framework of the widely used Simple Online and Real-time Tracking (SORT) algorithm, the proposed method tracks objects based on centroids and associations, making it well-suited for real-time vehicle tracking applications. One of the most compelling strengths of this unsupervised clustering method is its successful integration with the widely recognized Simple Online and Real-time Tracking (SORT) framework. Worth noting is that this method employs tracking by point, a concept distinct from the commonly used Intersection over Union (IoU) method in SORT. The method demonstrates impressive performance on datasets collected from three intersections. It outperforms the traditional bounding box method in tracking accuracy, achieving higher Multiple Object Tracking Accuracy (MOTA) and lower False Positive (FP) and False Negative (FN) rates. The promising results showcase the potential of the proposed method for real-world implementations in traffic management and safety assessment. The integration of the SORT framework and the concept of tracking by point, along with an unsupervised clustering method with the innovative 2D data structure converted from 3D LiDAR data after background filtering, presents a promising approach for optimizing vehicle tracking in roadside LiDAR applications. The study's significant performance improvements highlight the potential for real-time implementations. With continued research and refinement, the proposed method holds great promise for enhancing vehicle tracking accuracy and contributing to safer and more efficient traffic management systems for road users. | |
| dc.format.mimetype | Application/pdf | |
| dc.identifier.uri | https://hdl.handle.net/2346/97360 | |
| dc.language.iso | en | |
| dc.rights.availability | Access is not restricted. | |
| dc.subject | Track by point | |
| dc.subject | Simple Online and Real-time Tracking (SORT) | |
| dc.subject | Vehicle Trajectory | |
| dc.subject | Roadside LiDAR | |
| dc.subject | Background filtering | |
| dc.subject | Clustering | |
| dc.subject | Transportation management | |
| dc.title | Online vehicle trajectory extraction based on LiDAR data | |
| dc.type | Dissertation | |
| thesis.degree.department | Civil, Environmental and Construction Engineering | |
| thesis.degree.discipline | Civil Engineering | |
| thesis.degree.grantor | Texas Tech University | |
| thesis.degree.level | Doctoral | |
| thesis.degree.name | Doctor of Philosophy |