Dynamic Targets Occupancy Status Detection Utilizing mmWave Radar Sensor and Ensemble Machine Learning

Abstract

Rapid advancements in communication technologies in the Internet of Things (IoT) domain have had an impact on the application of positioning technology across multiple domains. Although there have been numerous fully fledged approaches for detection and localization in outdoor scenarios, due to high path loss and shadowing, these are insufficiently accurate in indoor scenarios. The primary enabler of various healthcare and safety applications is the precise sensing and localization of targets. A cost-effective approach with little maintenance is crucial for the development of such reliable systems. To address such sensing and localization challenges in indoor scenarios, we propose a novel dynamic target detection technique based on an ensembled convolutional neural network (CNN) classifier. An AWR1843 Radar sensor is used to collect data corresponding to dynamic targets in indoor scenarios. The range of each moving target in the room is estimated using point cloud data extracted from the received signal. An ensemble-based 1-D CNN classifier is used to analyze the data. To model the ensemble classifier, we used three CNN classifiers. The performances of the state-of-the-art classifiers considered in the comparison varied between 44$\%$ and 95$\%$ in terms of accuracy. In contrast, the proposed system attained an accuracy of 97.65$\%$ during training and 96.47$\%$ during testing and outperformed the state-of-the-art approaches.

Keywords

• Classsification
• convolutional neural network (CNN)
• dynamic target detection
• ensemble learning
• localization
• mmWave radar