Computer Science and Engineering

Permanent URI for this collection

https://hdl.handle.net/10315/28572

Browse

Now showing 1 - 20 of 110

Open Access
3D Modelling for Improved Visual Traffic Analytics
(2018-08-27) Soto, Eduardo R Corral; Elder, James H.
Advanced Traffic Management Systems utilize diverse types of sensor networks with the goal of improving mobility and safety of transportation systems. These systems require information about the state of the traffic configuration, including volume, vehicle speed, density, and incidents, which are useful in applications such as urban planning, collision avoidance systems, and emergency vehicle notification systems, to name a few. Sensing technologies are an important part of Advanced Traffic Management Systems that enable the estimation of the traffic state. Inductive Loop Detectors are often used to sense vehicles on highway roads. Although this technology has proven to be effective, it has limitations. Their installation and replacement cost is high and causes traffic disruptions, and their sensing modality provides very limited information about the vehicles being sensed. No vehicle appearance information is available. Traffic camera networks are also used in advanced traffic monitoring centers where the cameras are controlled by a remote operator. The amount of visual information provided by such cameras can be overwhelmingly large, which may cause the operators to miss important traffic events happening in the field. This dissertation focuses on visual traffic surveillance for Advanced Traffic Management Systems. The focus is on the research and development of computer vision algorithms that contribute to the automation of highway traffic analytics systems that require estimates of traffic volume and density. This dissertation makes three contributions: The first contribution is an integrated vision surveillance system called 3DTown, where cameras installed at a university campus together with algorithms are used to produce vehicle and pedestrian detections to augment a 3D model of the university with dynamic information from the scene. A second major contribution is a technique for extracting road lines from highway images that are used to estimate the tilt angle and the focal length of the camera. This technique is useful when the operator changes the camera pose. The third major contribution is a method to automatically extract the active road lanes and model the vehicles in 3D to improve the vehicle count estimation by individuating 2D segments of imaged vehicles that have been merged due to occlusions.
Open Access
A General FOFE-net Framework for Simple and Effective Question Answering over Knowledge Bases
(2019-11-22) Wu, Dekun; Jiang, Hui
Question answering over knowledge base (KB-QA) has recently become a popular research topic in NLP. One of the popular ways to solve the KBQA problem is to make use of a pipeline of several NLP modules, including entity discovery and linking (EDL) and relation detection. Recent success on KBQA task usually involves complex network structures with sophisticated heuristics. Inspired by a previous work that builds a strong KBQA baseline, we propose a simple but general neural model composed of fixed-size ordinally forgetting encoding (FOFE) and deep neural networks, called FOFE-net to solve KB-QA problem at different stages. For evaluation, we use two popular KB-QA datasets, SimpleQuestions, WebQSP, and our newly created dataset, FreebaseQA. The experimental results show that FOFE-net performs well on KBQA subtasks, entity discovery and linking (EDL) and relation detection, and in turn pushing overall KB-QA system to achieve strong results on all the datasets.
Open Access
A Message Transfer Framework for Enhanced Reliability in Delay-and Disruption-Tolerant Networks
(2016-09-20) Yasmeen, Farzana; Nguyen, Uyen T.
Many infrastructure-less networks require quick, ad hoc deployment and the ability to deliver messages even if no instantaneous end-to-end path can be found. Such networks include large-scale disaster recovery networks, mobile sensor networks for ecological monitoring, ocean sensor networks, people networks, vehicular networks and projects for connectivity in developing regions such as TIER (Technology and Infrastructure for Emerging Regions). These types of networks can be realized with delay-and disruption-tolerant network (DTN) technology. Generally, messages in DTNs are transferred hop-by-hop toward the destination in an overlay above the transport layer called the ''bundle layer''. Unlike mobile ad hoc networks (MANETs), DTNs can tolerate disruption on end-to-end paths by taking advantage of temporal links emerging between nodes as nodes move in the network. Intermediate nodes store messages before forwarding opportunities become available. A series of encounters (i.e., coming within mutual transmission range) among different nodes will eventually deliver the message to the desired destination. The message delivery performance (such as delivery ratio and delay) in a DTN highly depends on time elapsed between encounters (inter-contact time) and the time two nodes remain in each others communication range once a contact is established (contact-duration). As messages are forwarded opportunistically among nodes, it is important to have sufficient contact opportunities in the network for faster, more reliable delivery of messages. In this thesis, we propose a simple yet efficient method for increasing DTN performance by increasing the contact duration of encountered nodes (i.e., mobile devices). Our proposed sticky transfer framework and protocol enable nodes in DTNs to collect neighbors' information, evaluate their movement patterns and amounts of data to transfer in order to make decisions of whether to ''stick'' with a neighbor to complete the necessary data transfers. Nodes intelligently negotiate sticky transfer parameters such as stick duration, mobility speed and movement directions based on user preferences and collected information. The sticky transfer framework can be combined with any DTN routing protocol to improve its performance. Our simulation results show that the proposed framework can improve the message delivery ratio by up to 38% and the end-to-end message transfer delay by up to 36%.
Open Access
A Short and Readable Proof of Cut Elimination for Two 1st Order Modal Logics
(2016-11-25) Gao, Feng; Tourlakis, George
Since 1960s, logicians, philosophers, AI people have cast eyes on modal logic. Among various modal logic systems, propositional provability logic which was established by Godel modeling provability in axiomatic Peano Arithmetic (PA) was the most striking application for mathematicians. After Godel, researchers gradually explored the predicate case in provability logic. However, the most natural application QGL for predicate provability logic is not able to admit cut elimination. Recently, a potential candidate for the predicate provability logic ML3 and its precursors BM and M3 introduced by Toulakis,Kibedi, Schwartz dedicated that A is always closed. Although ML3, BM and M3 are cut free, the cut elimination proof with the unfriendly nested induction of high multiplicity is difficult to understand. In this thesis, I will show a cut elimination proof for all (Gentzenisations) of BM, M3 and ML3, with much more readable inductions of lower multiplicity.
Open Access
A Study of Colour Rendering in the In-Camera Imaging Pipeline
(2020-05-11) Karaimer, Hakki Can; Brown, Michael S.
Consumer cameras such as digital single-lens reflex camera (DSLR) and smartphone cameras have onboard hardware that applies a series of processing steps to transform the initial captured raw sensor image to the final output image that is provided to the user. These processing steps collectively make up the in-camera image processing pipeline. This dissertation aims to study the processing steps related to colour rendering which can be categorized into two stages. The first stage is to convert an image's sensor-specific raw colour space to a device-independent perceptual colour space. The second stage is to further process the image into a display-referred colour space and includes photo-finishing routines to make the image appear visually pleasing to a human. This dissertation makes four contributions towards the study of camera colour rendering. The first contribution is the development of a software-based research platform that closely emulates the in-camera image processing pipeline hardware. This platform allows the examination of the various image states of the captured image as it is processed from the sensor response to the final display output. Our second contribution is to demonstrate the advantage of having access to intermediate image states within the in-camera pipeline that provide more accurate colourimetric consistency among multiple cameras. Our third contribution is to analyze the current colourimetric method used by consumer cameras and to propose a modification that is able to improve its colour accuracy. Our fourth contribution is to describe how to customize a camera imaging pipeline using machine vision cameras to produce high-quality perceptual images for dermatological applications. The dissertation concludes with a summary and future directions.
Open Access
A Study on Deep Learning: Training, Models and Applications
(2018-03-01) Pan, Hengyue; Jiang, Hui
In the past few years, deep learning has become a very important research field that has attracted a lot of research interests, attributing to the development of the computational hardware like high performance GPUs, training deep models, such as fully-connected deep neural networks (DNNs) and convolutional neural networks (CNNs), from scratch becomes practical, and using well-trained deep models to deal with real-world large scale problems also becomes possible. This dissertation mainly focuses on three important problems in deep learning, i.e., training algorithm, computational models and applications, and provides several methods to improve the performances of different deep learning methods. The first method is a DNN training algorithm called Annealed Gradient Descent (AGD). This dissertation presents a theoretical analysis on the convergence properties and learning speed of AGD to show its benefits. Experimental results have shown that AGD yields comparable performance as SGD but it can significantly expedite training of DNNs in big data sets. Secondly, this dissertation proposes to apply a novel model, namely Hybrid Orthogonal Projection and Estimation (HOPE), to CNNs. HOPE can be viewed as a hybrid model to combine feature extraction with mixture models. The experimental results have shown that HOPE layers can significantly improve the performance of CNNs in the image classification tasks. The third proposed method is to apply CNNs to image saliency detection. In this approach, a gradient descent method is used to iteratively modify the input images based on pixel-wise gradients to reduce a pre-defined cost function. Moreover, SLIC superpixels and low level saliency features are applied to smooth and refine the saliency maps. Experimental results have shown that the proposed methods can generate high-quality salience maps. The last method is also for image saliency detection. However, this method is based on Generative Adversarial Network (GAN). Different from GAN, the proposed method uses fully supervised learning to learn G-Network and D-Network. Therefore, it is called Supervised Adversarial Network (SAN). Moreover, SAN introduces a different G-Network and conv-comparison layers to further improve the saliency performance. Experimental results show that the SAN model can also generate state-of-the-art saliency maps for complicate images.
Open Access
Achieving Adaptation Through Live Virtual Machine Migration in Two-Tier Clouds
(2015-08-28) Lu, Hong Bin; Litoiu, Marin
This thesis presents a model-driven approach for application deployment and management in two-tier heterogeneous cloud environments. For application deployment, we introduce the architecture, the services and the domain specific language that abstract common features of multi-cloud deployments. By leveraging the architecture and the language, application deployers author a deployment model that captures the high-level structure of the application. The deployment model is then translated into deployment workflows on specific clouds. As a use case, we introduce a live VM migration framework that maintains the application quality of services through VM migrations across two tier-clouds. The proposed framework can monitor the performance of the applications and their underlying infrastructure and plan and executes VM migrations to eliminate hotspots in a datacenter. We evaluate both the application deployment architecture and the live migration on public clouds.
Open Access
ACT-R Based Models For Learning Interactive Layouts
(2015-01-26) Das, Arindam; Stuerzlinger, Wolfgang
This dissertation presents research on learning of interactive layouts. I develop two models based on a theory of cognition known as ACT-R (Adaptive Control of Thought–Rational). I validate them against experimental data collected by other researchers. The first model is a simulation model that emulates the transition from novice to expert level in text entry. The model transcribes the presented English letters on a traditional phone keypad. It predicts the non-movement time to copy a pre-cued letter. It explains the visual exploration strategy that a user may employ in the novice to expert continuum. The second model is a closed-form model that accounts for the combined effect of practice, decay, proactive interference and mental effort on task completion time while practicing target acquisition on an interactive layout. The model can quantitatively compare a set of layouts in terms of the mental effort expended to learn them. My first model provides insight into how much practice is needed by a learner to progress from novice to expert level for an interactive layout. My second model provides insight into how effortful is it to learn a layout relative to other layouts.
Open Access
Active Control of Camera Parameters and Algorithm Selection for Object Detection
(2017-07-27) Wu, Yulong; Tsotsos, John K.
In this thesis, we quantitatively investigate the effect of camera parameters, shutter speed and voltage gain, on the performance of several popular object detection algorithms, under various illumination conditions. Our experimental results indicate a significant difference in sensitivity of the evaluated algorithms to these camera parameters. Based on the experimental benchmark results, a novel active control of camera parameters method and an algorithm selection extension are proposed. In empirical evaluation, our active control approach outperforms the conventional auto-exposure method for most algorithms. Also, the proposed algorithm selection extension has demonstrated the capability of selecting a proper tuple, in order to deal with varying light conditions.
Open Access
Adaptive Momentum for Neural Network Optimization
(2020-05-11) Rashidi, Zana; An, Aijun
In this thesis, we develop a novel and efficient algorithm for optimizing neural networks inspired by a recently proposed geodesic optimization algorithm. Our algorithm, which we call Stochastic Geodesic Optimization (SGeO), utilizes an adaptive coefficient on top of Polyaks Heavy Ball method effectively controlling the amount of weight put on the previous update to the parameters based on the change of direction in the optimization path. Experimental results on strongly convex functions with Lipschitz gradients and deep Autoencoder benchmarks show that SGeO reaches lower errors than established first-order methods and competes well with lower or similar errors to a recent second-order method called K-FAC (Kronecker-Factored Approximate Curvature). We also incorporate Nesterov style lookahead gradient into our algorithm (SGeO-N) and observe notable improvements. We believe that our research will open up new directions for high-dimensional neural network optimization where combining the efficiency of first-order methods and the effectiveness of second-order methods proves a promising avenue to explore.
Open Access
An Arm-Mounted Accelerometer and Gyro-Based 3D Control System
(2016-11-25) Young, Thomas Shih Teh; MacKenzie, I. Scott
This thesis examines the performance of a wearable accelerometer/gyroscope-based system for capturing arm motions in 3D. Two experiments conforming to ISO 9241-9 specifications for non-keyboard input devices were performed. The first, modeled after the Fitts' law paradigm described in ISO 9241-9, utilized the wearable system to control a telemanipulator compared with joystick control and the user's arm. The throughputs were 5.54 bits/s, 0.74 bits/s and 0.80 bits/s, respectively. The second experiment utilized the wearable system to control a cursor in a 3D fish-tank virtual reality setup. The participants performed a 3D Fitts' law task with three selection methods: button clicks, dwell, and a twist gesture. Error rates were 6.82 %, 0.00% and 3.59 % respectively. Throughput ranged from 0.8 to 1.0 bits/s. The thesis includes detailed analyses on lag and other issues that present user interface challenges for systems that employ human-mounted sensor inputs to control a telemanipulator apparatus.
Open Access
An Evaluation of Saliency and Its Limits
(2019-11-22) Wloka, Calden Frank; Tsotsos, John K.
The field of computational saliency modelling has its origins in psychophysical studies of visual search and low-level attention, but over the years has heavily shifted focus to performance-based model development and benchmarking. This dissertation examines the current state of saliency research from the perspective of its relationship to human visual attention, and presents research along three different but complementary avenues: a critical examination of the metrics used to measure saliency model performance, a software library intended to facilitate the exploration of saliency model applications outside of standard benchmarks, and a novel model of fixation control that extends fixation prediction beyond a static saliency map to an explicit prediction of an ordered sequence of saccades. The examination of metrics provides a more direct window into algorithm spatial bias than competing methods, as well as presents evidence that spatial considerations cannot be completely isolated from stimulus appearance when accounting for human fixation locations. Experimentation over psychophysical stimuli reveals that many of the most recent models, all which achieve high benchmark performance for fixation prediction, fail to identify salient targets in basic feature search, more complex singleton search, and search asymmetries, suggesting an overemphasis on the specific performance benchmarks that are widely used in saliency modelling research and a need for more diverse evaluation. Further experiments are performed to test how different saliency algorithms predict fixations across space and time, finding a consistent spatiotemporal pattern of saliency prediction across almost all tested algorithms. The fixation control model outperforms competing methods at saccade sequence prediction according to a number of trajectory-based metrics, and produces qualitatively more human-like fixation traces than those sampled from static maps. The results of these studies together suggest that the role of saliency should not be viewed in isolation, but rather as a component of a larger visual attention system, and this work provides a number of tools and techniques that will facilitate further understanding of visual attention.
Open Access
Analytically Defined Spatiotemporal ConvNets for Spacetime Image Understanding
(2020-08-11) Hadji, Isma; Wildes, Richard
This dissertation introduces a novel hierarchical spatiotemporal orientation representation for spacetime image analysis. This representation is designed to combine the benefits of the multilayer architecture of Convolutional Networks (ConvNets) and a more controlled approach to spacetime analysis. A distinguishing aspect of the approach is that unlike most contemporary convolutional networks no learning is involved; rather, all design decisions are specified analytically with theoretical motivations. This approach makes it possible to understand what information is being extracted at each stage and layer of processing as well as to minimize heuristic choices in design. Another key aspect of the network is its recurrent nature, whereby the output of each layer of processing feeds back to the input. The multilayer architecture that results systematically reveals hierarchical image structure in terms of multiscale, multiorientation properties of visual spacetime. To illustrate the utility of the proposed research, the designed networks has been tested on two spacetime image understanding tasks, dynamic texture recognition and video object segmentation. Further, the role of learning in the context of the proposed analytic approach to network design is systematically explored, thereby yielding a promising hybrid architecture. Finally, a new, large scale dynamic texture dataset is introduced and used for evaluation.
Open Access
Analyzing Human-Building Interactions in Virtual Environments Using Crowd Simulations
(2020-11-13) Usman, Muhammad; Faloutsos, Petros
This research explores the relationship between human-occupancy and environment designs by means of human behavior simulations. Predicting and analyzing user-related factors during environment designing is of vital importance. Traditional Computer-Aided Design (CAD) and Building Information Modeling (BIM) tools mostly represent geometric and semantic aspects of environment components (e.g., walls, pillars, doors, ramps, and floors). They often ignore the impact that an environment layout produces on its occupants and their movements. In recent efforts to analyze human social and spatial behaviors in buildings, researchers have started using crowd simulation techniques for dynamic analysis of urban and indoor environments. These analyses assist the designers in analyzing crowd-related factors in their designs and generating human-aware environments. This dissertation focuses on developing interactive solutions to perform spatial analytics that can quantify the dynamics of human-building interactions using crowd simulations in the virtual and built-environments. Partially, this dissertation aims to make these dynamic crowd analytics solutions available to designers either directly within mainstream environment design pipelines or as cross-platform simulation services, enabling users to seamlessly simulate, analyze, and incorporate human-centric dynamics into their design workflows.
Open Access
Approximate Parallel High Utility Itemset Mining
(2016-09-20) Chen, Yan; An, Aijun
High utility itemset mining discovers itemsets whose utility is above a given threshold, where utilities measure the importance of itemsets. In high utility itemset mining, memory and time performance limitations cause scalability issues, when the dataset is very large. In this thesis, the problem is addressed by proposing a distributed parallel algorithm, PHUI-Miner, and a sampling strategy, which can be used either separately or simultaneously. PHUI-Miner parallelizes the state-of-the-art high utility itemset mining algorithm HUI-Miner. The sampling strategy investigates the required sample size of a dataset, in order to achieve a given accuracy. We also propose an approach combining sampling with PHUI-Miner, which provides better time performance. In our experiments, we show that PHUI-Miner has high performance and outperforms the state-of-the-art non-parallel algorithm. The sampling strategy achieves accuracies much higher than the guarantee. Extensive experiments are also conducted to compare the time performance of PHUI-Miner with and without sampling.
Open Access
Automatic Extraction of Closed Contours Bounding Salient Objects: New Algorithms and Evaluation Methods
(2015-08-28) Movahedi, Vida; Elder, James
The problem under consideration in this dissertation is achieving salient object segmentation of natural images by means of probabilistic contour grouping. The goal is to extract the simple closed contour bounding the salient object in a given image. The method proposed here falls in the Contour Grouping category, searching for the optimal grouping of boundary entities to form an object contour. Our first contribution is to provide both a ground truth dataset and a performance measure for empirical evaluation of salient object segmentation methods. Our Salient Object Dataset (SOD) provides ground truth boundaries of salient objects perceived by humans in natural images. We also psychophysically evaluated 5 distinct performance measures that have been used in the literature and showed that a measure based upon minimal contour mappings is most sensitive to shape irregularities and most consistent with human judgements. In fact, the Contour Mapping measure is as predictive of human judgements as human subjects are of each other. Contour grouping methods often rely on Gestalt cues locally defined on pairs of oriented features. Accurate integration of these local cues with global cues is a challenge. A second major contribution of this dissertation is a novel, effective method for combining local and global cues. A third major contribution in this dissertation is a novel method based on Principal Component Analysis for promoting diversity among contour hypotheses, leading to substantial improvements in grouping performance. To further improve the performance, a multiscale implementation of this method has been studied. A fourth contribution in this dissertation is studying the effect of the multiscale prior on the performance and analysing the method for combining the results obtained in different resolutions. Our final contribution is comparing the performance of univariate distribution models for local cues used by our method with the use of a multivariate mixture model for their joint distribution. We obtain slight improvement by the mixture models. The proposed method has been evaluated and compared with four other state-of-the-art grouping methods, showing considerably better performance on the SOD ground truth dataset.
Open Access
Automatic Speech Recognition Using Deep Neural Networks: New Possibilities
(2015-08-28) Abdel-Hamid, Ossama Abdel-Hamid Mohamed; Jiang, Hui
Recently, automatic speech recognition (ASR) systems that use deep neural networks (DNNs) for acoustic modeling have attracted huge research interest. This is due to the recent results that have significantly raised the state of the art performance of ASR systems. This dissertation proposes a number of new methods to improve the state of the art ASR performance by exploiting the power of DNNs. The first method exploits domain knowledge in designing a special neural network (NN) structure called a convolutional neural network (CNN). This dissertation proposes to use the CNN in a way that applies convolution and pooling operations along frequency to handle frequency variations that commonly happen due to speaker and pronunciation differences in speech signals. Moreover, a new CNN structure called limited weight sharing is proposed to better suit special spectral characteristics of speech signals. Our experimental results have shown that the use of a CNN leads to 6-9% relative reduction in error rate. The second proposed method deals with speaker variations in a more explicit way through using a new speaker code based adaptation. This method adapts the speech acoustic model to a new speaker by learning a suitable speaker representation based on a small amount of adaptation data from each target speaker. This method alleviates the need to modify any model parameters as is done with other commonly used adaptation methods for neural networks. This greatly reduces the number of parameters to estimate during adaptation; hence, it allows rapid speaker adaptation. The third proposed method aims to handle the temporal structure within speech segments by using a deep segmental neural network (DSNN). The DSNN model alleviates the need to use an HMM model as it directly models the posterior probability of the label sequence. Moreover, a segment-aware NN structure has been proposed. It is able to model the dependency among speech frames within each segment and performs better than the conventional frame based DNNs. Experimental results show that the proposed DSNN can significantly improve recognition performance as compared with the conventional frame based models.
Open Access
Autonomous Robots in Dynamic Indoor Environments: Localization and Person-Following
(2018-05-28) Sahdev, Raghavender; Tsotsos, John K.
Autonomous social robots have many tasks that they need to address such as localization, mapping, navigation, person following, place recognition, etc. In this thesis we focus on two key components required for the navigation of autonomous robots namely, person following behaviour and localization in dynamic human environments. We propose three novel approaches to address these components; two approaches for person following and one for indoor localization. A convolutional neural networks based approach and an Ada-boost based approach are developed for person following. We demonstrate the results by showing the tracking accuracy over time for this behaviour. For the localization task, we propose a novel approach which can act as a wrapper for traditional visual odometry based approaches to improve the localization accuracy in dynamic human environments. We evaluate this approach by showing how the performance varies with increasing number of dynamic agents present in the scene. This thesis provides qualitative and quantitative evaluations for each of the approaches proposed and show that we perform better than the current approaches.
Open Access
Biolocomotion Detection in Videos
(2020-05-11) Kang, Soo Min; Wildes, Richard
Animals locomote for various reasons: to search for food, to find suitable habitat, to pursue prey, to escape from predators, or to seek a mate. The grand scale of biodiversity contributes to the great locomotory design and mode diversity. In this dissertation, the locomotion of general biological species is referred to as biolocomotion. The goal of this dissertation is to develop a computational approach to detect biolocomotion in any unprocessed video. The ways biological entities locomote through an environment are extremely diverse. Various creatures make use of legs, wings, fins, and other means to move through the world. Significantly, the motion exhibited by the body parts to navigate through an environment can be modelled by a combination of an overall positional advance with an overlaid asymmetric oscillatory pattern, a distinctive signature that tends to be absent in non-biological objects in locomotion. In this dissertation, this key trait of positional advance with asymmetric oscillation along with differences in an object's common motion (extrinsic motion) and localized motion of its parts (intrinsic motion) is exploited to detect biolocomotion. In particular, a computational algorithm is developed to measure the presence of these traits in tracked objects to determine if they correspond to a biological entity in locomotion. An alternative algorithm, based on generic handcrafted features combined with learning is assembled out of components from allied areas of investigation, also is presented as a basis of comparison to the main proposed algorithm. A novel biolocomotion dataset encompassing a wide range of moving biological and non-biological objects in natural settings is provided. Additionally, biolocomotion annotations to an extant camouflage animals dataset also is provided. Quantitative results indicate that the proposed algorithm considerably outperforms the alternative approach, supporting the hypothesis that biolocomotion can be detected reliably based on its distinct signature of positional advance with asymmetric oscillation and extrinsic/intrinsic motion dissimilarity.
Open Access
Biomechanical Locomotion Heterogeneity in Synthetic Crowds
(2020-05-11) Haworth, Michael Brandon; Faloutsos, Petros
Synthetic crowd simulation combines rule sets at different conceptual layers to represent the dynamic nature of crowds while adhering to basic principles of human steering, such as collision avoidance and goal completion. In this dissertation, I explore synthetic crowd simulation at the steering layer using a critical approach to define the central theme of the work, the impact of model representation and agent diversity in crowds. At the steering layer, simulated agents make regular decisions, or actions, related to steering which are often responsible for the emergent behaviours found in the macro-scale crowd. Because of this bottom-up impact of a steering model's defining rule-set, I postulate that biomechanics and diverse biomechanics may alter the outcomes of dynamic synthetic-crowds-based outcomes. This would mean that an assumption of normativity and/or homogeneity among simulated agents and their mobility would provide an inaccurate representation of a scenario. If these results are then used to make real world decisions, say via policy or design, then those populations not represented in the simulated scenario may experience a lack of representation in the actualization of those decisions. A focused literature review shows that applications of both biomechanics and diverse locomotion representation at this layer of modelling are very narrow and often not present. I respond to the narrowness of this representation by addressing both biomechanics and heterogeneity separately. To address the question of performance and importance of locomotion biomechanics in crowd simulation, I use a large scale comparative approach. The industry standard synthetic crowd models are tested under a battery of benchmarks derived from prior work in comparative analysis of synthetic crowds as well as new scenarios derived from built environments. To address the question of the importance of heterogeneity in locomotion biomechanics, I define tiers of impact in the multi-agent crowds model at the steering layer--from the action space, to the agent space, to the crowds space. To this end, additional models and layers are developed to address the modelling and application of heterogeneous locomotion biomechanics in synthetic crowds. The results of both studies form a research arc which shows that the biomechanics in steering models provides important fidelity in several applications and that heterogeneity in the model of locomotion biomechanics directly impacts both qualitative and quantitative synthetic crowds outcomes. As well, systems, approaches, and pitfalls regarding the analysis of steering model and human mobility diversity are described.