[Paper Review] Prediction-based Online Trajectory Compression

Arlei Silva, Ramya Raghavendra+2 more|arXiv (Cornell University)|Jan 23, 2016

Data Management and Algorithms27 references3 citations

TL;DR

ONTRAC is a novel online trajectory compression framework that suppresses redundant updates to trajectory databases using prediction models for spatial (Markov model) and temporal (QP-EM) components. It achieves up to 21× higher compression for travel-times and improves database scalability by an order of magnitude, even with 4-minute update delays, outperforming state-of-the-art offline methods.

ABSTRACT

Recent spatio-temporal data applications, such as car-shar\-ing and smart cities, impose new challenges regarding the scalability and timeliness of data processing systems. Trajectory compression is a promising approach for scaling up spatio-temporal databases. However, existing techniques fail to address the online setting, in which a compressed version of a trajectory stream has to be maintained over time. In this paper, we introduce ONTRAC, a new framework for map-matched online trajectory compression. ONTRAC learns prediction models for suppressing updates to a trajectory database using training data. Two prediction schemes are proposed, one for road segments via a Markov model and another for travel-times by combining Quadratic Programming and Expectation Maximization. Experiments show that ONTRAC outperforms the state-of-the-art offline technique even when long update delays (4 mininutes) are allowed and achieves up to 21 times higher compression ratio for travel-times. Moreover, our approach increases database scalability by up to one order of magnitude.

Motivation & Objective

• Address the scalability and timeliness challenges in real-time spatio-temporal data processing for applications like smart cities and ride-sharing.
• Overcome the limitations of existing offline trajectory compression techniques that cannot handle online, streaming trajectory data with low latency.
• Develop a framework that maintains accurate, compressed trajectory representations in real time while supporting efficient query processing.
• Enable efficient database operations by suppressing predictable updates using learned models for road segments and travel-times.
• Ensure compatibility with existing relational DBMS by designing a query-compatible compressed data schema with partial decompression support.

Proposed method

• Use a Markov model to predict the next road segment in a trajectory based on recent history, suppressing likely updates from being stored.
• Apply Quadratic Programming and Expectation Maximization to learn a Gaussian-based travel-time prediction model for temporal compression.
• Model temporal trajectory inference as a maximum-likelihood estimation problem to handle sparse and noisy GPS updates.
• Store only prediction errors (residuals) and prediction models, enabling full trajectory reconstruction at query time.
• Support partial decompression during queries to minimize performance overhead while maintaining accuracy.
• Integrate the compressed data schema with standard relational databases to enable indexing and efficient query processing.

Experimental results

Research questions

• RQ1 Can prediction-based suppression of trajectory updates improve online compression performance in real-time systems?
• RQ2 How does the compression ratio of an online prediction-based framework compare to state-of-the-art offline methods under realistic delay constraints?
• RQ3 To what extent can prediction models for road segments and travel-times reduce the number of database inserts in online trajectory processing?
• RQ4 What is the performance overhead of query processing when using partial decompression in the compressed trajectory framework?
• RQ5 Can the framework scale database operations by reducing the volume of incoming trajectory updates without sacrificing query accuracy?

Key findings

• ONTRAC achieves up to 21 times higher compression ratio for travel-time data compared to the state-of-the-art offline method, even with 4-minute update delays.
• The framework increases the database insert rate by up to 11.7 times (for Beijing dataset) compared to the uncompressed baseline.
• ONTRAC-PT (partial decompression) incurs only 25–38% overhead over the full uncompressed system, making it practical for real-time workloads.
• The system improves database scalability by up to one order of magnitude due to reduced data volume and insert frequency.
• The Markov model for road segment prediction and QP-EM for travel-time modeling enable accurate suppression of 95% of updates in real-world trajectories.
• Even with delayed updates, ONTRAC outperforms offline methods in compression ratio, demonstrating its effectiveness in online settings.