[Paper Review] Intrinsic Point of Interest Discovery from Trajectory Data

Matthew Piekenbrock, Derek Doran|arXiv (Cornell University)|Dec 14, 2017

Data Management and Algorithms34 references3 citations

TL;DR

This paper proposes a parameter-free framework for discovering intrinsic points of interest (POIs) directly from trajectory data using hierarchical density estimation, leveraging theoretical consistency results for nonparametric clustering. It achieves superior fidelity in POI detection across diverse trajectory datasets without manual tuning, outperforming state-of-the-art methods under thousands of parameter settings.

ABSTRACT

This paper presents a framework for intrinsic point of interest discovery from trajectory databases. Intrinsic points of interest are regions of a geospatial area innately defined by the spatial and temporal aspects of trajectory data, and can be of varying size, shape, and resolution. Any trajectory database exhibits such points of interest, and hence are intrinsic, as compared to most other point of interest definitions which are said to be extrinsic, as they require trajectory metadata, external knowledge about the region the trajectories are observed, or other application-specific information. Spatial and temporal aspects are qualities of any trajectory database, making the framework applicable to data from any domain and of any resolution. The framework is developed under recent developments on the consistency of nonparametric hierarchical density estimators and enables the possibility of formal statistical inference and evaluation over such intrinsic points of interest. Comparisons of the POIs uncovered by the framework in synthetic truth data to thousands of parameter settings for common POI discovery methods show a marked improvement in fidelity without the need to tune any parameters by hand.

Motivation & Objective

• To address the limitations of extrinsic POI definitions that rely on external gazetteers or application-specific metadata.
• To develop a data-driven, intrinsic POI discovery framework applicable to any geospatial trajectory dataset, regardless of resolution or domain.
• To enable automatic, parameter-free POI extraction using theoretical guarantees from hierarchical density estimation.
• To improve fidelity and usability in POI detection by eliminating the need for heuristic or ad hoc parameter tuning.
• To support exploratory analysis in complex environments like cities or buildings where POI count and structure are unknown in advance.

Proposed method

• The framework uses a relative surplus level (RSL) estimator based on the relative excess mass functional to identify density-based clusters in trajectory data.
• It applies hierarchical nearest-neighbor density estimation to construct a cluster tree, enabling multiscale detection of intrinsic POIs.
• The method selects POIs using a thresholding strategy on the RSL, with α ≥ √2 to ensure asymptotic consistency under theoretical conditions.
• The approach is grounded in recent minimax analysis of hierarchical density superlevel set estimators, ensuring statistical rigor.
• The framework is implemented in SUMO with extensions for pedestrian movement in indoor environments, enabling realistic simulation.
• Validation uses reproducible synthetic trajectory data across diverse geographic settings, comparing against multiple SOTA POI discovery algorithms.

Experimental results

Research questions

• RQ1 Can intrinsic POIs be reliably discovered from trajectory data alone, without relying on external gazetteers or metadata?
• RQ2 How does a theoretically grounded, parameter-free POI discovery framework compare to heuristic, parameter-sensitive methods in terms of detection fidelity?
• RQ3 To what extent does the proposed method maintain consistency and accuracy across varying trajectory data resolutions and spatial scales?
• RQ4 Can the framework support exploratory analysis in complex urban or indoor environments where POI structure is unknown?
• RQ5 What is the impact of theoretical consistency in density estimation on practical POI detection performance?

Key findings

• The proposed framework achieves significantly higher fidelity in POI detection compared to thousands of parameter settings of state-of-the-art methods, without any manual tuning.
• The method demonstrates robust performance across diverse synthetic trajectory datasets, including both pedestrian and vehicle movements in varied geographic contexts.
• The use of RSL with α ≥ √2 ensures theoretical consistency, avoiding the exponential k-dependence on dimensionality seen in other approaches.
• The framework is fully reproducible and open-source, with simulation data, validation code, and implementation available online.
• The approach outperforms existing methods such as DBSCAN, OPTICS, and HDBSCAN in detecting meaningful, spatially coherent POIs from raw trajectory data.
• The framework enables reliable, automated POI discovery in exploratory settings such as post-event urban analysis, where POI count and structure are a priori unknown.