Abstract
In a variety of fields, including security and search and rescue, it is vital to detect and locate living people behind walls. To address this challenge, a range of sensors are employed for detecting the presence of living people. One of the sensors that is used for this purpose is UWB Sensor. In this paper, we investigate feature extraction techniques for the detection of living people behind walls using data acquired with a UWB sensor. Experimental results demonstrate the effectiveness of different feature extraction approaches includes statistical features, time domain and frequency domain features besides machine learning algorithms. Comparative analysis of model performance and the impact of varying feature importance is provided. Notably, the top 5 features identified are primarily statistical features, highlighting their significant contribution to model accuracy. As the selected feature set expands to include top 10 important features, and beyond, a mix of statistical, time domain, and frequency domain features is incorporated. This comprehensive approach underscores the importance of leveraging a diverse range of feature types to optimize model performance. However, the subsequent addition of features beyond may introduce noise or redundancy and affect the model performance.