Spiking Neural Networks for People Counting Based on FMCW Radar

Abstract

This paper presents an innovative method for indoor people counting using Spiking Neural Networks (SNNs) exclusively with radar data, effectively addressing privacy concerns associated with camera-based systems. When compared to established neural network-based algorithms like VGG16 (F1-Score=0.8003), ResNet50 (F1-Score=0.6669), and MobileNet (F1-Score=0.7258), the SNN (F1-Score=0.8284) exhibits superior performance in the task of counting people indoors. Furthermore, the energy efficiency of the SNN is a strong advantage, particularly for hardware deployment. This characteristic not only reduces computational demands but also aligns with the increasing emphasis on energy conservation in modern technology. The combination of privacy preservation, accuracy, and energy efficiency positions the SNN as a promising choice for a wide range of real-world applications, including security, transportation, and smart spaces, offering a comprehensive solution for people counting in indoor environments.

Keywords

• FMCW radar
• people counting
• spiking neural network