Abstract:In large venues like shopping malls and airports, knowledge on the indoor populations fuels applications such as business analytics, venue management, and safety control. In this work, we provide means of modeling populations in partitions of indoor space offline and of monitoring indoor populations continuously, by using indoor positioning data. However, the low-sampling rates of indoor positioning render the data temporally and spatially sparse, which in turn renders the offline capture of indoor populations challenging. It is even more challenging to continuously monitor indoor populations, as positioning data may be missing or not ready yet at the current moment. To address these challenges, we first enable probabilistic modeling of populations in indoor space partitions as Normal distributions. Based on that, we propose two learning-based estimators for on-the-fly prediction of population distributions. Leveraging the prediction-based schemes, we provide a unified continuous query processing framework for a type of query that enables continuous monitoring of populated partitions. The framework encompasses caching and result validity mechanisms to reduce cost and maintain monitoring effectiveness. Extensive experiments on two real data sets show that the proposed estimators are able to outperform the state-of-the-art alternatives and that the query processing framework is effective and efficient.

What problem does this paper attempt to address?

This paper attempts to solve the problem of how to use sparse indoor positioning data to model and continuously monitor indoor crowds in large - scale indoor places (such as shopping malls and airports). Specifically, the paper focuses on the following two main issues: 1. **Historical indoor crowd modeling**: How to accurately estimate the number of people in an indoor area (sub - area) at a certain moment in the past through indoor positioning data with a low sampling rate. 2. **Real - time indoor crowd monitoring**: How to continuously monitor the crowd distribution in each sub - area indoors in real - time, even when the positioning data is missing or not ready at some moments. ### Detailed Explanation #### 1. Historical indoor crowd modeling Since indoor positioning data usually has a low sampling rate, the data is very sparse in both time and space, which makes it very difficult to directly count the historical number of people in a sub - area from the trajectory data. For this reason, the paper proposes a probability - based method, modeling the number of people in each sub - area during a certain period as a normal distribution \( N(\mu, \sigma^2) \), where: - \(\mu\) is the mean, representing the average number of people in the sub - area. - \(\sigma^2\) is the variance, representing the uncertainty of the number of people. The specific steps are as follows: - **Path inference**: For all possible paths between two consecutively reported positions, calculate the probability of each path, and infer the probability of an object passing through a sub - area according to these paths. - **Monte Carlo sampling**: Use the Monte Carlo method to randomly sample within the possible time range to determine the probability of an object in a sub - area. Finally, through multiple samplings and statistics, the crowd distribution in each sub - area during a certain period can be estimated. #### 2. Real - time indoor crowd monitoring Real - time monitoring is more challenging because it is necessary to predict the future crowd distribution under incomplete data. For this reason, the paper proposes two deep - learning - based estimators: - **Single - way Estimator (SE)**: Only predict for one sub - area, capturing the time - dependence of its historical population. - **Multi - way Estimator (ME)**: Simultaneously predict the population distributions of multiple sub - areas, considering the spatial dependence between different sub - areas. Both of these estimators use a multi - task learning framework to jointly predict the mean and variance of the normal distribution. In order to improve the query efficiency, the paper also designs a unified continuous query processing framework, including a feature sequence caching mechanism and a result validity mechanism, to reduce the cost of real - time prediction and maintain the effectiveness of monitoring. ### Summary The main contributions of the paper are: - Proposing a probability method based on the normal distribution to model historical indoor crowds. - Designing two neural network estimators for real - time prediction of the crowd distribution in indoor sub - areas. - Constructing a unified continuous monitoring framework to support efficient and accurate real - time crowd monitoring. These methods not only improve the accuracy of indoor crowd modeling and monitoring, but also have broad application prospects in practical applications, such as business analysis, safety management, etc.

Modeling and Monitoring of Indoor Populations using Sparse Positioning Data (Extension)

Modeling Real-Time Human Mobility Based on Mobile Phone and Transportation Data Fusion

Tracking the Dynamic Distribution of People in Indoor Space with Noisy Partitioning Sensors.

In Search of Indoor Dense Regions: an Approach Using Indoor Positioning Data

Population Mapping with Multisensor Remote Sensing Images and Point-Of-Interest Data.

Elm Meets Urban Computing: Ensemble Urban Data For Smart City Application

Detection of Indoor High-Density Crowds via Wi-Fi Tracking Data

Finding Most Popular Indoor Semantic Locations Using Uncertain Mobility Data.

RFID-Based Indoor Spatial Query Evaluation with Bayesian Filtering Techniques

Context-aware Real-Time Population Estimation for Metropolis

Probabilistic human mobility model in indoor environment

Monocular Occupancy Prediction for Scalable Indoor Scenes

Revealing Urban Dynamics by Learning Online and Offline Behaviours Together

Understanding Urban Dynamics Via State-Sharing Hidden Markov Model

Modeling occupancy distribution in large spaces with multi-feature classification algorithm

Continuous social distance monitoring in indoor space

Characterizing Large-Scale Population's Indoor Spatio-Temporal Interactive Behaviors.

Data Imputation for Sparse Radio Maps in Indoor Positioning (Extended Version)

A System of Monitoring and Analyzing Human Indoor Mobility and Air Quality

General Geospatial Inference with a Population Dynamics Foundation Model

Fine-grained spatiotemporal estimation of tourism flows leveraging cross-video collaborative perception