|
Road-based map matching and route interpolation on large-scale GPS data |
|---|---|
| รหัสดีโอไอ | |
| Title | Road-based map matching and route interpolation on large-scale GPS data |
| Creator | Athispat Phongampai |
| Contributor | Apichon Witayangkurn, Advisor |
| Publisher | Thammasat University |
| Publication Year | 2568 |
| Keyword | GPS trajectory, Apache spark, Map matching, Hidden markov model, Road networks, Route interpolation |
| Abstract | The analysis of mobility patterns in public transportation has gained significant attention alongside the increasing trend of urbanization, as evidenced by the substantial volume of Global Positioning System (GPS) trajectories collected. These data offer major benefits to vehicle navigation systems, traffic planning, and pollution monitoring. However, raw GPS data are unsuitable for direct use owing to intrinsic reliability limitations, necessitating map-matching techniques, particularly hidden Markov model (HMM)-based approaches, to align GPS trajectories with road networks. Although distributed computing frameworks enhance scalability, many existing solutions remain restricted within urban-scale deployments and often overlook road continuity in datasets with low sampling rates.This thesis presents a Hidden Markov Model (HMM)-based map-matching algorithm that incorporates vehicle heading, prioritizes assignments that remain on the current road segment, and utilizes an adaptive path length cache to enhance the processing of large-scale probe data, achieving improved accuracy. The algorithm effectively balances accuracy with computational performance by using Apache Spark for distributed processing, which enables efficient management of large datasets. Additionally, it includes a route interpolation mechanism to fill gaps in network coverage resulting from sparse measurement intervals. Moreover, stream processing capabilities have been implemented using Apache Kafka to facilitate the handling of real-world data types generated at relatively high speeds.In evaluating the results, we utilized trajectory, grid, and infrastructure-based approaches to assess the overall accuracy of the trajectory and the algorithm's performance across various road types, including how it handles transitions through bridges and tunnels. Experimental results from real-world taxi probe data in Thailand demonstrate the effectiveness of the proposed approach, achieving an average matching accuracy of 99.48% for vehicle-based evaluations and 99.56% for grid-based evaluations. Additionally, the algorithm significantly optimizes computational resources, reducing total processing time by up to 6.2 times compared to traditional HMM-based approaches. This method provides a scalable, accurate, and continuous map-matching performance, making it highly suitable for large-scale mobility analysis and transportation applications. |
