Identification and Prediction of Large Pedestrian Flow in Urban Areas Based on a Hybrid Detection Approach

Kaisheng Zhang, Mei Wang, Bangyang Wei, Daniel SunSustainability 2016 · DOI

TL;DR

This paper introduces a novel hybrid approach to identify and predict large pedestrian flows in urban areas by integrating data from mobile phone operators and base stations. This method leverages a multi-layered detection system, from coarse cell ID matching to fine-grained Wi-Fi and video analysis, to create a more robust and accurate crowd monitoring system. The ultimate goal is to enable better overcrowding detection and early warning systems for improved public safety.

What Problem Does This Paper Address?

Rapid urbanization has led to increasingly dense populations in cities, making overcrowded situations a significant safety concern, as tragically highlighted by events like the Shanghai Stampede on New Year's Eve 2015. Traditional methods for pedestrian flow analysis, primarily video-based, struggle with scalability, cost, and robustness to environmental factors like lighting and occlusion, especially in large-scale outdoor environments. The challenge is to develop a system that can accurately and reliably monitor pedestrian density and predict potential overcrowding in real-time, using readily available data sources.

Methods

The paper proposes a Hybrid Pedestrian Detection Model (HPDM) with five distinct levels of detection, progressively increasing in accuracy and detail:

  • Level 1 (Cell ID Matching): Provides a coarse overview of pedestrian density using single base station signals, with an accuracy of 200-300m.
  • Level 2 (Multi-BS Trilateration): Leverages signals from multiple base stations and models like Log Distance Path Loss (LDPL) and Angle of Arrival (AOA)/Direction of Arrival (DOA) to achieve accuracy of less than 100m.
  • Level 3 (Fingerprint Localization): Employs a probabilistic fingerprinting method to achieve pedestrian detection radii of around 10m.
  • Level 4 (Sensor Fusion): Integrates additional data sources like Wi-Fi, Bluetooth, and RFID to further enhance accuracy.
  • Level 5 (Video & Sensor Integration): Incorporates video processing at key locations and accelerator sensor data for trajectory prediction, representing the highest level of detail.

A key innovation is the use of temporal correlation of RSS (Received Signal Strength) for improved localization accuracy. This involves constructing intermediate matrices from RSS measurements and calculating correlation matrices to refine location estimates, especially in scenarios where raw data might be incomplete. For prediction, Gaussian Process (GP) models are used for both temporal and spatial data, allowing for regression and recovery of missed values, and enabling prediction of overcrowding events.

Datasets

The study utilizes real-time network data from a major mobile phone operator in China (China Telecom). This dataset includes:

  • Signaling and application records from 3,384,521 active users across 65,482 active cells over 20 days in January 2016.
  • Data types include Calling Detailed Records (CDR), Exchange Detailed Records (XDR), and User Field Detailed Records (UFDR), encompassing information like call details, internet usage, and application characteristics.

This dataset is described as being collected with industry best practices, and researchers signed privacy commitments.

Key Insights and Ideas

  • Leveraging Passive Data: The approach effectively utilizes passive localization by tapping into existing mobile network infrastructure, making it less intrusive and more scalable than active sensing methods.
  • Multi-Layered Detection: The hierarchical structure of HPDM allows for flexibility, providing a broad overview with lower accuracy methods and detailed insights with higher accuracy ones, catering to different monitoring needs.
  • Temporal Correlation is Key: Incorporating the temporal dynamics of RSS signals significantly improves localization accuracy, especially when dealing with sparse or noisy data. This is a crucial insight for robust positioning in dynamic environments.
  • Gaussian Processes for Prediction: GP models prove effective for both modeling complex spatial-temporal pedestrian flow patterns and for predicting future overcrowding scenarios, offering a probabilistic approach to early warning.
  • Real-World Data Validation: The use of a large-scale, real-world dataset from millions of users in a major urban center (Shanghai CBD) lends significant credibility to the proposed methods.

Key Results

  • The proposed hybrid approach significantly increases the utility and capacity of the mobile network.
  • The method enables the development of a more reasonable overcrowding detection and alert system.
  • Experiments demonstrated that the reliability of fingerprinting localization could be improved by up to 13% by utilizing temporal correlation.
  • The prediction results showed that the real data with added variance fell within the predicted upper and lower bounds, indicating the accuracy of the GP-based prediction model.
  • The modeling and regression capabilities of GP were shown to recover missed values, which is vital for real-time systems where data is often incomplete.

Limitations and Future Work

  • Data Scarcity for Multi-Scenario Analysis: The study acknowledges that multi-scenario analysis of the algorithm was limited due to data scarcity.
  • Spatial Properties Influence: Future work could further explore the influence of users' spatial properties on resource allocation.
  • Usage Rate Variability: The significant variability in user usage rates across different situations and times (as shown in Figure 5a) could be considered for constructing more accurate models.
  • Bandwidth Resource Allocation Details: Further research could delve into the finer details of bandwidth resource allocation and its impact.
  • Wider Applicability: Obtaining additional datasets would be beneficial to test the wider applicability of the proposed algorithm.

Summary updated

Abstract

Recently, population density has grown quickly with the increasing acceleration of urbanization. At the same time, overcrowded situations are more likely to occur in populous urban areas, increasing the risk of accidents. This paper proposes a synthetic approach to recognize and identify the large pedestrian flow. In particular, a hybrid pedestrian flow detection model was constructed by analyzing real data from major mobile phone operators in China, including information from smartphones and base stations (BS). With the hybrid model, the Log Distance Path Loss (LDPL) model was used to estimate the pedestrian density from raw network data, and retrieve information with the Gaussian Progress (GP) through supervised learning. Temporal-spatial prediction of the pedestrian data was carried out with Machine Learning (ML) approaches. Finally, a case study of a real Central Business District (CBD) scenario in Shanghai, China using records of millions of cell phone users was conducted. The results showed that the new approach significantly increases the utility and capacity of the mobile network. A more reasonable overcrowding detection and alert system can be developed to improve safety in subway lines and other hotspot landmark areas, such as the Bundle, People’s Square or Disneyland, where a large passenger flow generally exists.
BibTeX
@inproceedings{zhang2016identification,
  author = {Kaisheng Zhang and Mei Wang and Bangyang Wei and Daniel Sun},
  title = {{Identification and Prediction of Large Pedestrian Flow in Urban Areas Based on a Hybrid Detection Approach}},
  booktitle = {Sustainability},
  year = {2016},
  doi = {10.3390/su9010036},
}

Loading related content...

Figure