AI-Based Human Detection in Indoor Environments using IR-UWB Radar

This paper is built on a previously developed IRUWB radar-based detection approach for industrial indoor environments. The proposed signal processing steps resulted in range-Doppler maps of moving objects and humans, which are classified using machine/deep learning algorithms. We evaluate the performance of Support Vector Machines (SVM) and three convolutional neural network (CNN) architectures-AlexNet, GoogLeNet, and ResNet-18-across various signal-to-noise ratio (SNR) levels. Their effectiveness, assessed in terms of correct classification rate (CCR) and confusion matrix, is compared and discussed in the paper.