July 3, 2023

PoC system of pedestrian detection with super low energy consumption presented MVA 2023

PoC system of pedestrian detection with super low energy consumption presented MVA 2023
Our PoC system that can detect pedestrian with low energy consumption realized by event camera, SNN and neuromorphic hardware was presented at Internal Conference on Machine Vision Applications (MVA) 2023. In our paper, we showed that our Spiking Neural Network for pedestrian detection was ported onto a neuromorphic hardware named SpiNNaker.

Title of our paper is "Object Detection for Embedded Systems Using Tiny Spiking Neural Networks: Filtering Noise through Visual Attention. The points of this research are 1) small size SNN with visual attention layer, and 2) porting the SNN on a neuromorphic hardware named SpiNNaker (SpiNN-3). As the size of SNN architecture is small, more concretely, number of neuron is much less than SOTA, we achieved les energy consumption. At the same time, thanks to inserting attention layer, we improved also object detection performance even that architecture is compact. In addition to those, we confirmed that our SNN can execute on the SpiNNaker, which is one of representatives of neuromorphic hardware.

We are continue to focus on the technologies “SNN on neuromorphic hardware” to realize innovative and practical applications. For proceeding this project, we are seeking a partner especially neuromorphic hardware maker. If you are interested in this project, please contact us !!