Please use this identifier to cite or link to this item: http://hdl.handle.net/1959.13/933429
- Title
- Time series classification for analysing the impact of architectural design on pedestrian spatial behaviour
- Author/Creator
-
Jalalian, Arash
- Institution
- University of Newcastle. Faculty of Engineering & Built Environment, School of Electrical Engineering and Computer Science
- Description
- Research Doctorate - Doctor of Philosophy (PhD)
- Description
- Pedestrian spatial behaviour is defined as the pedestrians’ reaction to their immediate surroundings. Analysis of changes in this behaviour due to alternation in the environmental settings is an important facet of architectural and urban design. To measure the changes, human body dynamics, such as head position, gaze direction, movement direction, speed of movement, and trajectory can be employed. In this research the main purpose is to support architects and urban designers to better assess the impact of the spatial environment on the pedestrian’s behaviour in planned urban spaces. To this end, an analysis system is proposed to learn the patterns of behaviour observed in a simulated and real-world architectural space. The simulated environment is generated using the proposed pedestrian and urban models. The models provide important behavioural characteristics in a multi-agent-based simulation system. They support complex spatial interactions between agents and their environment, including agent-to-agent interactions, different spatial desires, and interpersonal distance. The simulated environment can be automatically generated using scanned line drawings of two-dimensional street maps or public spaces. In the simulation model, a variety of scenarios can be defined and modified by altering different parameters. Using the example of Wheeler Place in Newcastle (Australia), the experiments demonstrate how pedestrian behavioural characteristics can depend on selected abstract features in urban spaces. The characteristics are used in the analysis system to distinguish between different patterns of spatial behaviour. The analysis system consists of a proposed technique for sequential data classification where each data object may have different lengths. The new technique, called GDTW-P-SVMs, is a maximum margin method for the construction of classifiers with variable-length input series. It employs potential support vector machines (P-SVMs) and dynamic time warping (DTW) to waive the fixed-length restriction of feature vectors in standard support vector machines (SVMs). The new technique elaborates on the P-SVM kernel function, by utilising DTW to provide an elastic distance measure for the kernel function. Benchmarks for classification are performed with several real-world data sets from the UCR Time Series Classification/Clustering page, GeoLife trajectory data set, and UCI Machine Learning Repository. The data sets include data with both variable and fixed-length input series. The results show that the new method performs significantly better than the benchmarked standard classification methods. To learn patterns of spatial behaviour the proposed classification technique is employed with simulated and real-world characteristics. The characteristics are collected from Wheeler Place using the proposed simulation software and pedestrian tracking system. GDTW-P-SVMs classify patterns of behaviour using the whole sequence of data series as a single input to increase the classification performance. As a result, they can provide the highest classification accuracy using the simulated and real-world data sets, when compared with the other existing methods.
- Relation
- University of Newcastle Research Higher Degree Thesis
- Date
- 2012
- Keyword(s)
-
spatial behaviour analysis;
trajectory data analysis;
support vector machines;
dynamic time warping
- Resource Type
- thesis
- Rights
- Copyright 2012 Arash Jalalian
- Identifier
- http://hdl.handle.net/1959.13/933429
- Language
- eng
-
-
43 Visitors
81 Hits
23 Downloads
List Of Titles
