Search

Bayesian inference in its simplest forms is the act of moving from sample data to generalisations with a measurable degree of certainty through the utilisation of probability theory generally and Bayes' theorem specifically. Probabilistic Graphical Models (PGMs) on the other hand provide means to describe complex multivariate systems as joint probability distributions over multidimensional space yet pertaining a graph's unique ability to describe relations between system variables naturally and intuitively. Bayesian Networks (BNs) are a special type of PGMs that focuson modelling conditional joint probability distributions over multivariate domains, Dynamic Bayesian Networks (DBNs) are an extension of BNs to model conditionaljoint probability distributions over sequences of random variable spanning the time domain. The work in this dissertation investigates the potential and power of using Dynamic Bayesian Networks and Bayesian inference techniques for the sole purpose of reasoning over uncertainty in three challenging domains. These are the domains of human activity recognition and detection based multi-object tracking, where recognising human activities and tracking humans plays a fundamental role in the fields of machine learning and computer vision due to its potential in applications such human computer interaction and video surveillance systems, yet they are considered as very challenging time series classification problems, that involves the use of large streams of sensor data to enable action recognition and tracking. Another domain is anomaly detection which involves finding patterns of data that do not conform to what is normal, which can be very advantageous in critical system applications in fields such as network security and bank fraud detection. The first part of this dissertation will go through the theoretical background of Bayesian Networks and Dynamic Bayesian Networks, different aspects will be explored, from representation to parameter estimation and inference techniques. Next a probabilistic modelling and inferencing framework using DBNs and machine learning techniques is proposed, for classifying and predicting human activities from a temporal perspective. This model is first implemented on an aeroplane pilot flying activity domain, through the use of a software simulator. Next, a framework for anomaly detection through means of classification using DBNs is proposed. Afterwards this framework is extended for anomaly detection in the temporal domain using DBNs. These models are then implemented for detecting pilot errors, utilising the previous DBN pilot activity model along with different probabilistic inference techniques. The results are then evaluated and analysed. At the end, a novel Multiple Object Tracking (MOT) algorithm based on the Tracking-by-detection paradigm using DBNs is proposed. The tracker is tested using the MOT15, MOT16, MOT17 challenge benchmarks of video sequences. This algorithm is further enhanced by the introduction of additional affinity measures, in the form of a state of the art feature extractor built using a residual neural network and trained on the MARS person re-identification database. Finally, we improve the tracker performance by the use of a state of the art object detector (YOLOV5), custom trained on Google's Open Images dataset. Results are analysed and evaluated using the MOTChallenge Evaluation Kit, and then compared to other methods.

Due to economic and operational constraints, there is an increasing demand from aviation operators and training manufacturers to extract maximum training usage from the lower fidelity suite of flight simulators. It is possible to augment low-fidelity flight simulators to achieve equivalent performance compared to high-fidelity setups but at reduced cost and greater mobility. In particular for visual manoeuvres, the virtual reality technique of head-tracking amplification for virtual view control enables full field-of-regard access even with limited field-of-view displays. This research quantified the effects of this technique on piloting performance, workload and simulator sickness by applying it to a fixed-base, low-fidelity, low-cost flight simulator. In two separate simulator trials, participants had to land a simulated aircraft from a visual traffic circuit pattern whilst scanning for airborne traffic. Initially, a single augmented display was compared to the common triple display setup in front of the pilot. Starting from the base leg, pilots exhibited tighter turns closer to the desired ground track and were more actively conducting visual scans using the augmented display. This was followed up by a second experiment to quantify the scalability of augmentation towards larger displays and field of views. Task complexity was increased by starting the traffic pattern from the downwind leg. Triple displays in front of the pilot yielded the best compromise delivering flight performance and traffic detection scores just below the triple projectors but without an increase in track deviations and the pilots were also less prone to simulator sickness symptoms. This research demonstrated that head augmentation yields clear benefits of quick user adaptation, low-cost, ease of systems integration, together with the capability to negate the impact of display sizes yet without incurring significant penalties in workload and incurring simulator sickness. The impact of this research is that it facilitates future flight training solutions using this augmentation technique to meet budgetary and mobility requirements. This enables deployment of simulators in large numbers to deliver expanded mission rehearsal previously unattainable within this class of low-fidelity simulators, and with no restrictions for transfer to other training media.

Stewart Platforms have been used for simulation because of their strength and position accuracy. Simulators have remained large structures that contained internal replicas of airplane cockpits and/or driving seats. These large structures come at high prices, it would be beneficial for the training and gaming industry or markets to make smaller more affordable platforms that can perform at or near the same motion performance of large expensive platforms. An added benefit to include for the training community would be to add portability so the simulator can travel to a classroom setting. In this thesis, tools such as MATLAB and Mechanical Analysis are used to design a smaller, more affordable platform that is capable of similar performance capabilities to professional motion platforms.

This thesis describes the development of a fixed base flight simulator capable of hardware-in-the-loop testing of aviation electronics, or “avionics”. The system is composed of a pilot controlled 6 degree-of-freedom aircraft model, a flight display, and a hardware interface to an ARINC 429 avionics databus. Avionics can be connected to this databus for testing. These systems are combined in a modular fashion allowing easy re-configuration for use with various aircraft models, and testing of avionics using databuses other than ARINC 429. This modular design is achieved by using an aircraft model constructed in Matlab’s Simulink, and by relaying the model data to the display and ARINC 429 interface using the UDP network protocol.

Although the accident rate for military aviation has declined significantly from earlier decades, during the 1990's it reached a plateau. Human error in the cockpit still accounts for over 80% of the aircraft mishaps resulting in loss of life or over one million dollars in damage. Decision error has been a contributing factor for approximately 60% of these mishaps. The purpose of this research was to investigate aircrew process performance variables as predictors of decision-making outcomes. This study was modeled on elements of previous research in naturalistic decision making. Data were collected for cross-case analysis of the role experience plays in efficient decision strategy selection and use in an uncertain, dynamic high stakes environment. Multiple raters evaluated eight novice and eight experienced military aircrews at seven decision points in a 20-minute flight scenario conducted in a full motion flight simulator. Other raters independently rank ordered the quality of the final outcome. A comprehensive approach to collecting and analyzing data included: (1) development and use of a behaviorally-anchored assessment instrument, (2) use of a digitally integrated presentation of audio/video and flight data, and (3) development of context-specific analytical frameworks and models of observed behaviors and metacognitive processes. Results included inferential and descriptive statistics of process/outcome scores, instructor comments, excerpts of cockpit recordings, participant interviews, and field notes. The study findings were: (1) high individual and collective crew experience had a significant positive effect on process and outcome scores, (2) there was no statistically reliable difference in process scores between experience levels in the three procedurally-based events, (3) experienced crews performed better than novice crews in the four less structured events, (4) novice crews' process/outcome correlation did not approach significance, (5) a strong positive correlation of process/outcome scores was found for experienced crews in the two most challenging (i.e., unstructured) scenario events, (6) qualitative analysis revealed strong relationships between performance and crew interactions/attributes, and (7) in dynamic, time critical situations, the use of adaptive decision-making strategies led to better performance outcomes.