University of Twente Student Theses

Login

Leveraging LiDAR Data and Local Digital Twins Framework for Data-Driven Traffic Simulation

Wibisana, M.I. (2024) Leveraging LiDAR Data and Local Digital Twins Framework for Data-Driven Traffic Simulation.

This is the latest version of this item.

[img] PDF
6MB
Abstract:Urban planning, specifically in the field of transportation planning, is important for promoting economic and social activities within urban area. Globally, transportation planning faces significant challenges, including increment in traffic congestion, which affects over half of the urban areas. Despite temporary traffic density reductions during the COVID-19 pandemic, congestion remains a persistent issue, with average drivers losing considerable hours to traffic annually. Additionally, traffic accidents cause numerous fatalities and injuries worldwide each year, underscoring the need for improved traffic management strategies. Similarly, Bulgaria, particularly Sofia, face similar challenges, including high road fatality rates and significant congestion, driving the implementation of the Sustainable Urban Mobility Plan (SUMP) 2019-2035. This plan aims to digitalize city transport to enhance urban mobility. Accordingly, this research aims to develop an urban traffic simulation Digital Twin (DT) framework by utilizing detailed traffic data, primarily in the Open Serialization Format (OSEF), captured by LiDAR sensors. These sensors provide comprehensive information about real-world traffic conditions. By integrating this detailed data, the DT framework seeks to improve traffic simulations, enable traffic observation in the digital world and potentially reduce congestion. The study area chosen for this research is a busy intersection in Sofia near the Paradise Center Mall, the largest mall in Sofia, where the LiDAR sensors are located. This area is suitable for extensive traffic observation and analysis. The traffic simulation tool chosen for this research is Simulation of Urban MObility (SUMO) due to its versatile uses and analytical capabilities, as well as its open-source nature, which is well-suited for the DT framework. To develop the DT framework, the methods involve proper identifications, including issues inherent in the .osef dataset. It is found that there are challenges in the coordinate nature of the dataset and inherent classification that include a lot of unknown classes. Therefore, the method is first to transform the local coordinates of the .osef dataset concurrently with road segmentation of the intersection, followed by object type and trajectory type analysis, which is used as the base information for the reclassification process using Random Forest (RF) mode. This handles the issues of classification, including the existence of multiple classes in a single tracked object. Utilizing the PostGIS database as middleware to fetch enriched datasets for the traffic simulations, the framework demonstrates a successful attempt at running initial traffic simulations based on real-world traffic conditions. The findings in this research show that DT frameworks have shown that it is capable of integrating the .osef dataset into the traffic simulations while addressing issues such as the coordinate transformations, including the reclassification of unknown class. The RF model employed in this research able to predict the true label of the unknown class into a specific object type (e.g., car, truck, two-wheeler or person). The minimization of multiple class prediction is reduced by 76% with the 2nd tuned model. The DT framework itself is validated against real-world traffic survey data, revealing a strong alignment with an R-squared value of 0.97, indicating that 97% of the variability in the observed traffic counts is explained by the simulated counts. Additionally, the cosine similarity analysis for vehicle trajectories demonstrated high directional accuracy, with most cosine similarity values nearing perfection (one in scale), and the Euclidean distance confirming minimal positional deviation. The implications of this research are more prominent in the what-if scenario testing, where it shows that roundabout road network design seems to be the most suitable to reduce traffic congestion. This research recommends that for future work, it is better to focus on testing the framework scalability across different intersections and residential areas, improving the Random Forest model with more advanced machine learning or deep learning techniques, and extending the observation period to capture more comprehensive traffic patterns.
Item Type:Essay (Master)
Clients:
The Big Data for Smart Society Institute (GATE), Sofia, Bulgaria
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Subject:38 earth sciences, 50 technical science in general, 55 traffic technology, transport technology
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/101576
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page