EbESCNN: A Eye Binary Event Spiking Convolutional Neural Network Architecture for Event-Based Eye Tracking

This paper presents EbESCNN, a novel Spiking Convolutional Neural Network (SCNN) designed for near-eye pupil tracking using event-based camera data. Operating on binary event representations, the proposed architecture predicts pupil coordinates by employing the membrane potential of spiking neurons as a continuous output for regression. Evaluated on the 3ET+ dataset, the network achieves 86% accuracy within a 10-pixel margin, with a mean Euclidean error of 5.68 pixels, comparable to state-of-the-art methods. Despite long training times, the architecture demonstrates strong potential for efficient and accurate eye tracking in resource-constrained and neuromorphic environments.