Accidents inside an indoor area can result in serious injuries. Therefore, technology used to detect human activities inside an indoor area can support the elderly or patients with degenerative diseases such as dementia. This detection can monitor safety and prevent incidents. However, some technologies may be of concern in terms of privacy. There are several methods for monitoring accidents, such as, camera-based, wearable sensor-based, radio frequency signal-based, and ultra-wideband (UWB)-based technologies. In this paper, UWB-based technology was chosen because of its high accuracy and ability to provide precise distance measurements. We proposed a human posture and position detection system by integrating into UWB technology with machine learning (ML) models to improve accuracy in closed-room environments. UWB signals are collected and analyzed in the form of channel frequency response (CFR) using multiple ML algorithms, including XGBoost, random forest, Naïve Bayes, and artificial neural networks (ANNs), to classify six human activities and four zone localization. The proposed approach demonstrates promising results in posture recognition and indoor localization. The results show that ANNs provide the highest in terms of evaluation metrics, 0.8914 accuracy, and XGBoost achieves the highest macro-AUC-value of 0.9935.