Automatic Identification of the Social Functions of Areas of Interest (AOIs) Using the Standard Hour-Day-Spectrum Approach

Tong Zhou, Xintao Liu, Zhen Qian, Haoxuan Chen, Fei TaoISPRS Int. J. Geo Inf 2020 · DOI

TL;DR

This paper introduces a novel "hour-day-spectrum" (HDS) approach to automatically identify the social functions of Areas of Interest (AOIs) using taxi GPS trajectory data. By analyzing the temporal patterns of taxi pick-ups and drop-offs, the method achieves a high accuracy of 90.74% in classifying AOIs, offering a dynamic and fine-grained alternative to traditional methods.

What Problem Does This Paper Address?

Identifying the social functions of urban areas (like residential, commercial, or recreational zones) is crucial for urban planning and smart city initiatives. Existing methods often rely on static data like remote sensing images, which lack temporal detail and can only identify land use at a coarse block level. Alternatively, dynamic data sources like social media or traffic data can be noisy and lead to low accuracy or high uncertainty. This paper aims to bridge this gap by leveraging the rich temporal and spatial information embedded in taxi GPS trajectories to provide a more accurate and dynamic method for classifying urban AOIs.

Methods

The core of this research is the Hour-Day-Spectrum (HDS) approach. Here's a breakdown of the methodology:

  • Data Preprocessing: Taxi GPS trajectories are meticulously cleaned to remove spatial errors and noise. This involves a "double cleaning" process to ensure the accuracy of pick-up and drop-off points.
  • AOI Association: Buffer analysis and DBSCAN clustering are used to associate pick-up and drop-off points with specific AOIs, considering both closed and open entrances.
  • Hour-Day-Spectrum (HDS) Construction: For each AOI, the temporal patterns of taxi activity are analyzed over a 24-hour period, differentiating between weekdays and holidays, and between pick-up and drop-off events. This results in six types of HDS waveforms (total, holiday, and weekday for both pick-ups and drop-offs).
  • Standard HDS (SHDS) Training: Representative HDS waveforms from sampled AOIs are used to create "standard" HDS (SHDS) for each AOI type (e.g., residential, school, hospital, shopping mall). This involves sampling and interpolation to create a generalized spectral representation.
  • Social Function Identification: The k-Nearest Neighbors (KNN) algorithm is employed to classify unknown AOIs. The HDS of an unknown AOI is compared against the trained SHDSs using similarity metrics.
  • Similarity Metrics: The paper evaluates three similarity indicators: Cosine Similarity, Pearson Correlation Coefficient, and the Gaussian Kernel Function. The Gaussian Kernel Function, with a tuned parameter (2σ²=1.5), proved to be the most effective.

Datasets

  • Taxi Trajectory Data: The study utilized original taxi GPS trajectory data from approximately 1,400 taxis in Nantong, China, collected between September and October 2018. This data included license plate number, call sign, time, latitude, longitude, speed, direction, and passenger status.
  • AOI Data: AOI data for Nantong, China, was obtained using Amap API via web crawling. The selected AOIs included shopping malls, schools, hospitals, and residential districts.
  • Training Set: 50 AOI samples from the Chongchuan District were used to construct the standard HDSs.
  • Validation Set: 108 AOI samples from the Gangzha District were used for testing and validation.

Key Insights and Ideas

  • Taxi Trajectories as a Proxy for Human Behavior: The paper demonstrates that taxi pick-up and drop-off patterns are strong indicators of an AOI's social function, effectively acting as a proxy for human travel behavior and activity.
  • The Power of Temporal Patterns (HDS): By analyzing the "hour-day-spectrum," the study captures the unique temporal "fingerprints" of different AOI types, revealing distinct daily and weekly activity rhythms.
  • Dynamic and Fine-Grained Classification: Unlike static methods, the HDS approach allows for continuous updates and fine-grained identification of land use types, crucial for understanding evolving urban dynamics.
  • Object-Oriented Approach: The method is AOI-centric, allowing for integration with image interpretation and other methods to enhance identification accuracy.

Key Results

  • High Accuracy: The proposed method achieved an overall accuracy of 90.74% in identifying the social functions of AOIs.
  • Gaussian Kernel Function Superiority: Among the tested similarity metrics, the Gaussian Kernel Function demonstrated the best performance, achieving the highest accuracy.
  • Identification of Six HDS Types: The approach successfully generates six types of HDS waveforms (total, holiday, and weekday for pick-ups and drop-offs) that serve as effective interpretation indicators.
  • Validation: Experiments were conducted on 108 AOI samples, confirming the robustness of the method.

Limitations and Future Work

  • Mutual Interference between Adjacent AOIs: The study acknowledges that AOIs can exist at different hierarchical levels, and adjacent AOIs with significant functional differences (e.g., a large hospital next to a school) can lead to spectral information being misallocated, causing incorrect classifications. The authors suggest accumulating data over longer periods and using smaller buffer areas for future big data analysis to address this.
  • Newly Built or Abnormally Status AOIs: AOIs that are newly built or have unusual statuses (e.g., low occupancy, incomplete access) may not generate sufficient trajectory data for accurate spectral analysis.
  • Impact of Spatial Location: The regularity of travel behavior might be weaker at the edge of a city compared to its center.
  • City Size and AOI Richness: The study was conducted in Nantong, a smaller city with a less diverse range of AOI types. The authors suggest that applying the method to larger cities with more varied AOIs could yield even better results.
  • Enclosed AOIs: The method is particularly well-suited for cities where AOIs are often enclosed by walls and have limited entrances, which is common in China. Adapting it for cities with more open management modes might require modifications.
  • Future Work: The authors propose combining this method with image interpretation and other identification techniques for improved results. They also suggest exploring longer data accumulation periods and smaller buffer areas to mitigate issues with AOI interference.

Summary updated

BibTeX
@article{zhou2020automatic,
  author = {Tong Zhou and Xintao Liu and Zhen Qian and Haoxuan Chen and Fei Tao},
  title = {{Automatic Identification of the Social Functions of Areas of Interest (AOIs) Using the Standard Hour-Day-Spectrum Approach}},
  journal = {ISPRS Int. J. Geo Inf.},
  year = {2020},
  doi = {10.3390/ijgi9010007},
}
Figure