Ben Miled, Meriem; (2024) Efficient Bio-inspired System for Dynamic Obstacles Detection and Avoidance for Autonomous UAVs. Doctoral thesis (Ph.D), UCL (University College London).

Text Efficient_Bio_inspired_System_for_Dynamic_Obstacles_Detection_and_Avoidance_for_Autonomous_UAVs.pdf - Published Version Access restricted to UCL open access staff until 1 January 2026. Download (25MB)

Abstract

In this thesis, we explore the frontiers of bio-inspired dynamic obstacle detection for autonomous Unmanned Aerial Vehicles (UAVs) utilizing event cameras. These neuromorphic cameras, designed to mimic the retina’s ganglion cells, offer unparalleled temporal resolution at the microsecond level, producing a near continuous, asynchronous stream of data that can captures changes in light intensity with minimal motion blur. However, the data generated by these cameras, especially in outdoor environments, is significantly prone to noise caused by variation in lighting conditions. This research investigates the capabilities and limitations of current state-of-the-art, event-data-driven algorithms when applied to real outdoor scenarios with varying light conditions. We will propose a novel approach to denoising and detection of dynamic obstacles that draws inspiration from mammalian motion perception mechanisms. This innovative method aims to effectively filter out noise while accurately identifying dynamic obstacles, thereby addressing the primary challenges posed by the noise inherent in event camera data. Additionally, we develop a fusion platform that integrates various data sources, further enhancing the system’s ability to detect obstacles under diverse environmental conditions. Finally, we shall introduce a holistic algorithm based on spike neural networks (SNN), that eliminates the need for a separate denoising phase. This approach leverages the inherent advantages of spike-based processing to more closely mimic the human perception system. Through a combination of bio-inspired algorithms and advanced neural network models, this research contributes to the field by providing a robust framework for dynamic obstacle detection in UAVs. This framework not only improves the reliability and efficiency of autonomous flight in complex outdoor environments, but also paves the way for future advancement in neuromorphic vision systems and their application in robotics.

Download activity - last month

Export as JSONXMLCSV

Download activity - last 12 months

Export as CSVXMLJSON

Downloads by country - last 12 months

Export as CSVXMLJSON

Archive Staff Only

View Item