Event-driven neuromorphic vision enables energy-efficient visual place recognition
Abstract
Reliable visual place recognition (VPR) under dynamic real-world conditions is critical for autonomous robots, yet conventional deep networks remain limited by high computational and energy demands. Inspired by the mammalian navigation system, we introduce SpikeVPR, a bio-inspired neuromorphic framework that leverages event-based cameras and spiking neural networks (SNNs) to learn compact, invariant place descriptors from a limited number of exemplars, enabling robust VPR across severe changes in illumination, viewpoint, and appearance. SpikeVPR is trained end-to-end using surrogate gradient learning and incorporates EventDilation, a novel augmentation strategy enhancing robustness to speed and temporal variations. Across three challenging benchmarks (Brisbane-Event-VPR, NSAVP and NYC-Event-VPR), SpikeVPR matches the performance of state-of-the-art deep networks while reducing model size by 50 × and lowering estimated energy consumption by 30–250 × , based on an analytical energy model, enabling efficient real-time deployment on mobile and neuromorphic hardware. These results demonstrate that spike-based coding offers an efficient pathway toward robust VPR in complex, changing environments.
Related
- Recognition of auditory objects from continuous and ultra-sparse spike streams in neuromorphic devices
- PSMamba: Progressive self-supervised vision Mamba for plant disease recognition
- MambaSeed: A lightweight Mamba-based hybrid vision network for tomato leaf disease recognition
- Lightweight frequency-aware hybrid vision transformer for fine-grained plant disease recognition
- A modular open-source computer vision pipeline enables individual-level behavior analysis of group-housed pigs
BibTeX
@article{keime2026eventdriven,
title = {{Event-driven neuromorphic vision enables energy-efficient visual place recognition}},
author = {Geoffroy Keime and Nicolas Cuperlier and Benoit R Cottereau},
journal = {Neuromorphic Computing and Engineering},
year = {2026},
month = jul,
doi = {10.1088/2634-4386/ae8a77},
url = {https://hal.science/hal-05847260v1},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - Event-driven neuromorphic vision enables energy-efficient visual place recognition
AU - Geoffroy Keime
AU - Nicolas Cuperlier
AU - Benoit R Cottereau
PY - 2026
DA - 2026/07/28
JO - Neuromorphic Computing and Engineering
DO - 10.1088/2634-4386/ae8a77
UR - https://hal.science/hal-05847260v1
AB - Reliable visual place recognition (VPR) under dynamic real-world conditions is critical for autonomous robots, yet conventional deep networks remain limited by high computational and energy demands. Inspired by the mammalian navigation system, we introduce SpikeVPR, a bio-inspired neuromorphic framework that leverages event-based cameras and spiking neural networks (SNNs) to learn compact, invariant place descriptors from a limited number of exemplars, enabling robust VPR across severe changes in illumination, viewpoint, and appearance. SpikeVPR is trained end-to-end using surrogate gradient learning and incorporates EventDilation, a novel augmentation strategy enhancing robustness to speed and temporal variations. Across three challenging benchmarks (Brisbane-Event-VPR, NSAVP and NYC-Event-VPR), SpikeVPR matches the performance of state-of-the-art deep networks while reducing model size by 50 × and lowering estimated energy consumption by 30–250 × , based on an analytical energy model, enabling efficient real-time deployment on mobile and neuromorphic hardware. These results demonstrate that spike-based coding offers an efficient pathway toward robust VPR in complex, changing environments.
SN - 2634-4386
ER -