Development of a Computer Vision Framework for Improved Remotely Piloted Aircraft Operations

Abstract

This thesis proposes a computer vision framework to enable improved operations of Remotely Piloted Aircraft equipped with onboard image sensors. The main use of payload image sensors is to provide visual imagery data to the system for real-time or postprocessing applications; the application of an image quality metric and the ground sampling distance of the image sensor can be used to predict the performance of an image sensor in enabling the image classification task. This information is used to determine the mission-specific operational envelope of the aircraft, to ensure that visual data quality requirements are met. The application of a convolutional neural network for image processing is also presented. Finally, a vision-based positioning system is developed; it achieves an average position estimation difference of 18 cm compared to a commercially available indoor localization system and provides a position update rate at 12 Hz. viii 4.3

Near Infrared Imaging and Image Pre-Processing to Improve the Automatic Detection of Canada Geese

“…As shown in the results section, there were minimal geese detected in the 25 m altitude test image, but more detections were seen when the RPAS flew closer to the decoys. This is attributed to Johnson's criteria where the resolution of the object in the image is correlated to the number of pixels of the object (Ooi, 2019). The more pixel cycles there are within the object, the greater the chances of the object being recognized and identified.…”

Deep Learning and Synthetic Imagery for Migratory Bird Species Identification Using Drones

“…There are several successful cases of applications with uninhabited aerial systems used in wildlife surveying and monitoring (Wilson et al, 2022;Reckling et al, 2021;Rominger and Meyer, 2021;Schroder et al, 2021;Marzialetti et al, 2021), with many focusing in ornithology in particular (Ooi, 2020;Dulava et al, 2015;Valle and Scarton, 2019a;Corregidor-Castro and Valle, 2022;Bushaw et al, 2020). The use of UAS in ornithology is not a new concept, but recent advances in technology have made it possible to collect data with a higher quality and at a lower cost, leading to increased popularity.…”

“…Imagery data with live Canada geese collected from UAS was made available from prior work of one of the members of the project, Brendan Ooi (Ooi, 2020). The images were extracted from a short clip recorded in Carleton University's campus in summer of 2019, and were used for testing the classifier model.…”

“…In many ecological research drone missions, data is collected in the form of images and videos for the development of computer vision applications (Wilson et al, 2022;Corregidor-Castro and Valle, 2022), often employing deep learning models (Reckling et al, 2021;Rominger and Meyer, 2021;Richardson et al, 2021;Ooi, 2020). The success of deep learning applications with drone imagery is associated with the flexibility offered by lightweight aircraft carrying high resolution cameras that are capable of collecting large volumes of data from a bird's eye view perspective.…”

“…The collaboration between the Molson Foundation and Carleton University, initiated in 2019, have already yielded two Masters theses by Brendan Ooi (Ooi, 2020) and Jacqueline Szeto (Szeto, 2022) in Carleton University: the first, introducing a general framework for computer vision in UAS bird surveying and the latter using object detection models and near infrared imaging for locating birds. The work presented in this present thesis is a continuation of their work, and the prior contributions of each member in the team will be made clear in the following chapters.…”