Neuromorphic vision for optical inter- satellite link acquisition

Abstract

This thesis presents a complete FPGA implementation of a two-stage neuromorphic vision pipeline targeting the coarse acquisition and tracking stage of space-based laser communication terminals. Event cameras respond asynchronously per-pixel with microsecond-level latency and a photometric dynamic range of 120–160 dB, producing an event stream that separates a moving satellite beacon from a static star-field background. The first stage implements a Cache-Like Filter (CLF) algorithm achieving O(m+n) on-chip memory complexity for an m×n sensor, sustaining one-event-per-clock throughput at 125 MHz. The second stage implements a Boundary-Movement Multi-Object Tracker algorithm, performing all detection and tracking decisions entirely in hardware. The filter reduces the event stream sufficiently to enable stable object tracking within a five-tracker hardware budget, and precision above 0.96 is achievable across noise-to-signal ratios up to 2.0 under a controlled synthetic evaluation. The complete design occupies under 10% of available logic and block memory resources on a Xilinx Kria KV260.