Efficient Human Posture Detection Using mmWave Radar and Neural Networks on PYNQ-ZU SoC

This paper details the development of a lightweight Artificial Neural Network (ANN) model for posture classification utilizing mmWave radar technology, specifically employing the Texas Instrument’s IWR1843BOOST platform and implementing deep learning classifier on the PYNQ-ZU FPGA. The research uses a dataset of spatial coordinates collected from the radar system, with the goal of accurately classifying different postures in real-time scenarios. By harnessing the distinct advantages of mmWave radar, known for its reliable performance in varied environmental conditions, the model achieved a remarkable accuracy of 96.5%. The combination of the ANN architecture with FPGA processing not only boosts computational efficiency but also significantly lowers latency, making it ideal for embedded applications. Our results highlight the potential of mmWave radar paired with advanced machine learning methods for effective posture recognition, opening new avenues for applications in health monitoring, human-computer interaction, and smart environments.