2D and 3D crowd analysis and simulation using deep learning and big data analysis

Every future city will have a digital `twin' that consumes data from the physical city and generates predictions for its design, construction, and management. The centre-piece of this digital twin is its residents. This research project is to build such residents, a `digital twin' of crowds, by proposing the next-generation mathematical models and artificial intelligence algorithms. This will be achieved by combining machine learning with neuroscience, architectural design and crowd management. Today, it is expected that more than 6.7 billion people will aggregate in urban spaces by 2050, leading to megacities of 10 million inhabitants (United Nations). The research project will lead to a crowd-driven framework that can predict crowd motions, help design new spaces and improve existing spaces, to eliminate potential dangers, minimise discomfort and maximise efficiency, enabling planners and policymakers to meet the great challenges of fast urbanisation in the 21st century.

This project is to look into the fundamental crowd motions in different environments including indoor/outdoor scenarios. The goal of this project is to propose a series of new models and mathematical frameworks to capture the crowd motions for the purposes of analysis and simulation.

The research falls into the category of data-driven crowd analysis where data is intensively used for analysis as compared to traditional empirical modelling where concise mathematical models are made trying to capture the complex structure in the data. However, because of the high complexity, more data (big data) is needed and meanwhile corresponding algorithms and models with enough capacity for data consumption are to be developed. 

The project is currently accepting PhD applications every year.