Wi-Fi Sensing for Motion Detection using ESP-32 and Raspberry Pi 4: A Comparison

In recent studies, both the ESP-32 and Raspberry Pi 4 (RPi 4) have been widely used for capturing Channel State Information (CSI) in motion detection applications. However, to the best of our knowledge, there has been no direct comparison between the performance and reliability of these two popular platforms. In this paper, we present an in-depth analysis of the CSI data captured from both the ESP-32 and RPi 4 in motion detection. Specifically, we evaluate the efficacy of Deep Neural Networks (DNNs) trained on CSI data from each platform, investigating the accuracy, robustness, and reliability of motion detection systems. The highest performance achieved with the ESP-32 was 99.26% using a CNN model, while the best performance for the RPi 4 was 98.15% using an LSTM model. In terms of reliability, we analyze the confidence histogram, Brier score, and log cross-entropy loss. Our analysis demonstrates that, although both hardware devices yield similar accuracies in motion detection, models trained on CSI data with the ESP-32 show greater reliability compared to those trained on data from the RPi 4.