Event camera-based robot localisation and optical communication

Chen, Yixin 2026. Event camera-based robot localisation and optical communication. PhD Thesis, Cardiff University. Item availability restricted.

	PDF - Accepted Post-Print Version Restricted to Repository staff only until 10 November 2027 due to copyright restrictions. Download (4MB)
	PDF (Cardiff University Electronic Publication Form) - Supplemental Material Restricted to Repository staff only Download (1MB)

Abstract

Autonomous robots depend on visual information to determine where they are and acquire information needed for their tasks. Event cameras offer a different basis for these capabilities by reporting brightness changes asynchronously, with precise timing at individual pixels. However, this temporal precision does not directly provide the spatial context, metric support or temporal association required for localisation and communication. The central challenge is therefore to turn sparse, unevenly distributed changes into observations that can meaningfully support a robotic task. This thesis develops spatiotemporal methods for using event-camera measure ments in localisation, local motion inference and visual communication. Its cen tral approach is to make the spatial support, reliability and temporal relevance of observations explicit when they are incorporated into an estimator or decoder. For localisation, a quality-gated eventLiDAR point line graph connects event derived visual structure with metric depth and registration constraints. A shared observation interface preserves the support and origin of point and line measure ments, while selective factor admission and delayed landmark commitment govern their influence on the graph. This enables an investigation of how structural in formation can be retained alongside pose estimates when visual observations and depth support are uneven. For local motion inference, a rolling Gaussian Belief Propagation backend com bines partial normal- ow constraints through a hierarchy of local angular-velocity variables. The contribution adapts local Gaussian message passing to an evolving observation set: measurement quality, observation age and finite history jointly determine which information is retained and aggregated. This provides a means to examine the relationship between local evidence, graph organisation and estimation cost. The thesis also extends the use of event measurements to information transmitted by a display. A display-aware receiver combines spatial cell assignment with transition-triggered, stateful decoding to recover binary payloads. Its temporal gating accounts for the difference between scheduled symbols and the optical transitions observed from a consumer display, enabling a fixed-pose visual link for potential stop-and-read robotic interactions. Across these investigations, the thesis examines how spatial support and temporal relevance govern the use of event-camera measurements. The evaluations identify conditions for incorporating visual structure into localisation and reveal how observation retention and receiver timing affect inference and decoding. The resulting methods connect asynchronous brightness changes to geometric con straints, local probabilistic estimates and visual payloads, broadening the use of event cameras for robotic perception and information acquisition.

Item Type:	Thesis (PhD)
Date Type:	Completion
Status:	Unpublished
Schools:	Schools > Engineering
Uncontrolled Keywords:	1. Event cameras 2. Robotic localisation 3. Gaussian belief propagation 4. optical camera communication 5. Event-based vision 6. Event–LiDAR sensor fusion
Date of First Compliant Deposit:	10 November 2026
Last Modified:	10 Nov 2026 16:38
URI:	https://orca.cardiff.ac.uk/id/eprint/190800

Actions (repository staff only)

Edit Item