Please use this identifier to cite or link to this item:
Type: Thesis
Title: An Insect-Inspired Target Tracking Mechanism for Autonomous Vehicles
Author: Bagheri, Zahra
Issue Date: 2017
School/Discipline: School of Mechanical Engineering
Abstract: Target tracking is a complicated task from an engineering perspective, especially where targets are small and seen against complex natural environments. Due to the high demand for robust target tracking algorithms a great deal of research has focused on this area. However, most engineering solutions developed for this purpose are often unreliable in real world conditions or too computationally expensive to be used in real-time applications. While engineering methods try to solve the problem of target detection and tracking by using high resolution input images, fast processors, with typically computationally expensive methods, a quick glance at nature provides evidence that practical real world solutions for target tracking exist. Many animals track targets for predation, territorial or mating purposes and with millions of years of evolution behind them, it seems reasonable to assume that these solutions are highly efficient. For instance, despite their low resolution compound eyes and tiny brains, many flying insects have evolved superb abilities to track targets in visual clutter even in the presence of other distracting stimuli, such as swarms of prey and conspecifics. The accessibility of the dragonfly for stable electrophysiological recordings makes this insect an ideal and tractable model system for investigating the neuronal correlates for complex tasks such as target pursuit. Studies on dragonflies identified and characterized a set of neurons likely to mediate target detection and pursuit referred to as ‘small target motion detector’ (STMD) neurons. These neurons are selective for tiny targets, are velocity-tuned, contrast-sensitive and respond robustly to targets even against the motion of background. These neurons have shown several high-order properties which can contribute to the dragonfly’s ability to robustly pursue prey with over a 97% success rate. These include the recent electrophysiological observations of response ‘facilitation’ (a slow build-up of response to targets that move on long, continuous trajectories) and ‘selective attention’, a competitive mechanism that selects one target from alternatives. In this thesis, I adopted a bio-inspired approach to develop a solution for the problem of target tracking and pursuit. Directly inspired by recent physiological breakthroughs in understanding the insect brain, I developed a closed-loop target tracking system that uses an active saccadic gaze fixation strategy inspired by insect pursuit. First, I tested this model in virtual world simulations using MATLAB/Simulink. The results of these simulations show robust performance of this insect-inspired model, achieving high prey capture success even within complex background clutter, low contrast and high relative speed of pursued prey. Additionally, these results show that inclusion of facilitation not only substantially improves success for even short-duration pursuits, it also enhances the ability to ‘attend’ to one target in the presence of distracters. This inspect-inspired system has a relatively simple image processing strategy compared to state-of-the-art trackers developed recently for computer vision applications. Traditional machine vision approaches incorporate elaborations to handle challenges and non-idealities in the natural environments such as local flicker and illumination changes, and non-smooth and non-linear target trajectories. Therefore, the question arises as whether this insect inspired tracker can match their performance when given similar challenges? I investigated this question by testing both the efficacy and efficiency of this insect-inspired model in open-loop, using a widely-used set of videos recorded under natural conditions. I directly compared the performance of this model with several state-of-the-art engineering algorithms using the same hardware, software environment and stimuli. This insect-inspired model exhibits robust performance in tracking small moving targets even in very challenging natural scenarios, outperforming the best of the engineered approaches. Furthermore, it operates more efficiently compared to the other approaches, in some cases dramatically so. Computer vision literature traditionally test target tracking algorithms only in open-loop. However, one of the main purposes for developing these algorithms is implementation in real-time robotic applications. Therefore, it is still unclear how these algorithms might perform in closed-loop real-world applications where inclusion of sensors and actuators on a physical robot results in additional latency which can affect the stability of the feedback process. Additionally, studies show that animals interact with the target by changing eye or body movements, which then modulate the visual inputs underlying the detection and selection task (via closed-loop feedback). This active vision system may be a key to exploiting visual information by the simple insect brain for complex tasks such as target tracking. Therefore, I implemented this insect-inspired model along with insect active vision in a robotic platform. I tested this robotic implementation both in indoor and outdoor environments against different challenges which exist in real-world conditions such as vibration, illumination variation, and distracting stimuli. The experimental results show that the robotic implementation is capable of handling these challenges and robustly pursuing a target even in highly challenging scenarios.
Advisor: Cazzolato, Ben
Dissertation Note: Thesis (Ph.D.) -- University of Adelaide, School of Mechanical Engineering, 2017
Keywords: Bio-inspired vision
bio-inspired robot
target tracking
insect physiology
Provenance: This electronic version is made publicly available by the University of Adelaide in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. This thesis may incorporate third party material which has been used by the author pursuant to Fair Dealing exceptions. If you are the owner of any included third party copyright material you wish to be removed from this electronic version, please complete the take down form located at:
Appears in Collections:Research Theses

Files in This Item:
File Description SizeFormat 
Bagheri2017_PhD.pdf8.8 MBAdobe PDFView/Open

Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.