Crowd Analysis and simulation using deep learning and big data analysis
Lead Research Organisation:
University of Leeds
Department Name: Sch of Computing
Abstract
Motivation
More than half the world's population live in or around a city, leading to transportation systems often running at capacity, putting managers, operators and transport planners under pressure to ensure that overcrowding does not endanger the safety of passengers, and that vital transport links do not break down due to congestion. Understanding and modelling crowd behaviours is not only a necessity for daily crowd management and design of public spaces (airports, train stations, shopping districts), but also vital in planning events with many attendees such as the Olympics. Poor management may lead to tragic results. More recently, such concerns have been raised to the level of public safety and national security under terrorist attacks and catastrophic events. How crowds react to such events directly should influence the design and management of spaces. Crowd simulation, previously widely used for movies/games, has shown its effectiveness in simulating and predicting crowds, therefore the potential to be applied on the aforementioned problems. This has been noticed by the government. However, one breakthrough needs to be made in next-generation tools: realistic motions need to be modeled and rigorous and quantitative assessments need to be conducted. This is exactly the goals of the proposed research.
Aims and Objectives
The principal goal of this research programme is to build computational models of human crowd dynamics, by investigating fundamental issues in crowd motion, leveraging data of new forms and developing innovative machine learning methods. The outcome is a series of new mathematical models and algorithms for crowd analysis, simulation, and prediction.
Objectives
1 Build a statistical machine learning model for combined analysis of crowds in space, time and dynamics.
2 Build a simulation guidance model based on the analysis model to guide crowd simulation practice.
3 Propose an evaluation metric based on the analysis model to compare different simulation methods by comparing simulated data with real data.
Potential Applications
The first impact will be on the UK game and movie industry. The two industries combined contributed £8.8 billion UK GDP in 2014-2015, by more than 2,000 game companies and over 120k people directly or indirectly employed in film production nation-wide (Source: BFI and UKIE). The research outcome has the potential to help generate content that involved labourous manual work, thus greatly increasing the productivity and lower the costs. In addition, the market of crowd management in transportation is expected to reach $1,100m globally (Market Research Future 2017 report). Being able to analyse, simulate and predict crowd dynamics, especially in high-density areas, will enable space designers/managers and policy makers to lower costs, increase infrastructure capacity and avoid chaos. Finally, the market of Autonomous Vehicles is expected to be grown to £907bn globally and £28bn domestically by 2035, with 10k jobs rising from the technology only (Transport Systems Catapult Report 2017). However, the safety becomes a major concern. Using simulated crowds to test existing control algorithm and develop new ones has just been experimented. A realistic crowd simulator will provide greater reliability and generalisability for these tests.
More than half the world's population live in or around a city, leading to transportation systems often running at capacity, putting managers, operators and transport planners under pressure to ensure that overcrowding does not endanger the safety of passengers, and that vital transport links do not break down due to congestion. Understanding and modelling crowd behaviours is not only a necessity for daily crowd management and design of public spaces (airports, train stations, shopping districts), but also vital in planning events with many attendees such as the Olympics. Poor management may lead to tragic results. More recently, such concerns have been raised to the level of public safety and national security under terrorist attacks and catastrophic events. How crowds react to such events directly should influence the design and management of spaces. Crowd simulation, previously widely used for movies/games, has shown its effectiveness in simulating and predicting crowds, therefore the potential to be applied on the aforementioned problems. This has been noticed by the government. However, one breakthrough needs to be made in next-generation tools: realistic motions need to be modeled and rigorous and quantitative assessments need to be conducted. This is exactly the goals of the proposed research.
Aims and Objectives
The principal goal of this research programme is to build computational models of human crowd dynamics, by investigating fundamental issues in crowd motion, leveraging data of new forms and developing innovative machine learning methods. The outcome is a series of new mathematical models and algorithms for crowd analysis, simulation, and prediction.
Objectives
1 Build a statistical machine learning model for combined analysis of crowds in space, time and dynamics.
2 Build a simulation guidance model based on the analysis model to guide crowd simulation practice.
3 Propose an evaluation metric based on the analysis model to compare different simulation methods by comparing simulated data with real data.
Potential Applications
The first impact will be on the UK game and movie industry. The two industries combined contributed £8.8 billion UK GDP in 2014-2015, by more than 2,000 game companies and over 120k people directly or indirectly employed in film production nation-wide (Source: BFI and UKIE). The research outcome has the potential to help generate content that involved labourous manual work, thus greatly increasing the productivity and lower the costs. In addition, the market of crowd management in transportation is expected to reach $1,100m globally (Market Research Future 2017 report). Being able to analyse, simulate and predict crowd dynamics, especially in high-density areas, will enable space designers/managers and policy makers to lower costs, increase infrastructure capacity and avoid chaos. Finally, the market of Autonomous Vehicles is expected to be grown to £907bn globally and £28bn domestically by 2035, with 10k jobs rising from the technology only (Transport Systems Catapult Report 2017). However, the safety becomes a major concern. Using simulated crowds to test existing control algorithm and develop new ones has just been experimented. A realistic crowd simulator will provide greater reliability and generalisability for these tests.
