Answers are grounded in this paper and cite their sources.
Address-event representation (AER)-based imaging offers high-speed, high-dynamic-range vision by processing the pixel-level visual signals in the form of spatiotemporal, asynchronous spikes. The conventional AER systems rely on complex CMOS silicon spiking neuron circuits and can suffer from dense spike generation and transmission as well as high hardware overhead, limiting their applications, particularly for the edge scenarios. To address the challenges, we propose a spiking neuron design using the emerging threshold switching memristors and, by exploiting the threshold switching dynamics of the memristors, the neurons achieve neural spiking behavior with compact circuits and operations. The neurons, via signal summation and differencing, allow efficient sparse spike encoding of analog and digital signals, facilitating their use in lightweight AER imagers for edge applications. We demonstrate the feasibility by integrating the neuron array into an AER motion-tracking imager architecture and realizing motion tracking in a 1000 fps video, with >97% sparsity in neuron activation while preserving a 4.3% accuracy degradation in inferring the motion details, e.g. the trajectory and velocity. Our work provides a hardware-efficient solution for next-generation AER imaging strategy for edge motion tracking.
No takes yet. Share an insight, caveat, or question.
Liu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: