In many different insect species the highly conserved neuropil regions known as the central complex or central body complex have been shown to be important in behaviours such as locomotion, visual memory and courtship conditioning. The aim of this project is to generate accurate quantitative neuroanatomy of the central complex in the fruit fly Drosophila melanogaster. Much of the authoritative neuroanatomy of the fruit fly from past literature has been derived using Golgi stains, and in important cases these data are available only from 2D camera lucida drawings of the neurons and linguistic descriptions of connectivity. These cannot easily be mapped onto 3D template brains or compared directly to our own data. Using GAL4 driver and reporter constructs, some of the findings within these studies could be visualized using immunohistochemistry and confocal microscopy. A range of GAL4 driver lines were selected that particularly had prominent expression in the fan-shaped body. Images of brains from these lines were archived using a web-based 3D image stack archive developed for the sharing and backup of large confocal stacks. This is also the platform which we use to publish the data, so that other researchers can reuse this catalogue and compare their results directly. Each brain was annotated using desktop-based tools for labelling neuropil regions, locating landmarks in image stacks and tracing fine neuronal processes both manually and automatically. The development of the tracing and landmark annotation tools is described, and all of the tools used in this work are available as free software. In order to compare and aggregate these data, which are from many different brains, it is necessary to register each image stack onto some standard template brain. Although this is a well-studied problem in medical imaging, these high resolution scans of the central fly brain are unusual in a number of respects. The relative effectiveness of various methods currently available were tested on this data set. The best registrations were produced by a method that generates free-form deformations based on B-splines (the Computational Morphometry Toolkit), but for much faster registrations, the thin plate spline method based on manual landmarks may be sufficient. The annotated and registered data allows us to produce central complex template images and also files that accurately represent the possible central complex connectivity apparent in these images. One interesting result to arise from these efforts was evidence for a possible connection between the inferior region of the fan-shaped body and the beta lobe of the mushroom body which had previously been missed in these GAL4 lines. In addition, we can identify several connections which appear to be similar to those described in [Hanesch et al., 1989], the canonical paper on the architecture of the Drosophila melanogaster central complex, and describe for the first time their variation statistically. This registered data was also used to suggest a method for classifying layers of expression within the fan-shaped body.

Flying animals such as bats, birds, and moths are actively studied by researchers wanting to better understand these animals’ behavior and flight characteristics. Towards this goal, multi-view videos of flying animals have been recorded both in lab- oratory conditions and natural habitats. The analysis of these videos has shifted over time from manual inspection by scientists to more automated and quantitative approaches based on computer vision algorithms. This thesis describes a study on the largely unexplored problem of 3D pose estimation of flying animals in multi-view video data. This problem has received little attention in the computer vision community where few flying animal datasets exist. Additionally, published solutions from researchers in the natural sciences have not taken full advantage of advancements in computer vision research. This thesis addresses this gap by proposing three different approaches for 3D pose estimation of flying animals in multi-view video datasets, which evolve from successful pose estimation paradigms used in computer vision. The first approach models the appearance of a flying animal with a synthetic 3D graphics model and then uses a Markov Random Field to model 3D pose estimation over time as a single optimization problem. The second approach builds on the success of Pictorial Structures models and further improves them for the case where only a sparse set of landmarks are annotated in training data. The proposed approach first discovers parts from regions of the training images that are not annotated. The discovered parts are then used to generate more accurate appearance likelihood terms which in turn produce more accurate landmark localizations. The third approach takes advantage of the success of deep learning models and adapts existing deep architectures to perform landmark localization. Both the second and third approaches perform 3D pose estimation by first obtaining accurate localization of key landmarks in individual views, and then using calibrated cameras and camera geometry to reconstruct the 3D position of key landmarks. This thesis shows that the proposed algorithms generate first-of-a-kind and leading results on real world datasets of bats and moths, respectively. Furthermore, a variety of resources are made freely available to the public to further strengthen the connection between research communities.