Hardware Deployable Edge AI Solution for Posture Classification Using mmWave Radar and Low- Computational Machine Learning Model

• Published in: IEEE sensors journal Vol. 24; no. 16; pp. 26836 - 26844
• Main Authors: Pratap Singh, Yash, Gupta, Aham, Chaudhary, Devansh, Wajid, Mohd, Srivastava, Abhishek, Mahajan, Pranjal
• Format: Journal Article
• Language: English
• Published: New York IEEE 15.08.2024; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Accuracy; Artificial Intelligence (AI); Chirp; Classification; Continuous radiation; Datasets; Hardware; Machine learning; millimeter wave (mmWave); Millimeter wave communication; Millimeter waves; point-cloud images; Posture; posture classification; Privacy; Radar; Radar data; Radar imaging; Sensors; Spatial data; Three dimensional models; Three-dimensional displays; tiny machine learning
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3416390

Identifying correct human postures is crucial in areas, such as patient care, in hospitals. However, the traditional vision-based methods widely used for this purpose raise privacy concerns for the subject, and the other wearable sensor-based approaches are impractical for real-world scenarios. In this article, we propose a contactless, privacy-conscious, and memory-efficient posture classification system based on a millimeter-wave (mmWave) radar. This system utilizes 3-D point-cloud data captured using Texas Instrument's IWR1843BOOST frequency-modulated continuous-wave (FMCW) radar module to classify the posture of the subject. Two types of datasets are extracted from these radar data: 1) image dataset derived from the isometric view of the point-cloud data and 2) spatial coordinates dataset also extracted from the point-cloud data. A low-computational tiny machine learning (TinyML) model is employed on the datasets for efficient implementation on embedded hardware, Raspberry Pi 3 B+. The proposed model's parameters were quantized to 8 bits (int8), which accurately classify four postures, i.e., standing, sitting, lying, and bending, with an accuracy of 98.97% for the image data. However, to make it more computationally efficient, the int8 quantized TinyML model was trained on the spatial coordinates dataset, giving an accuracy of 96.12%. This highlights the efficiency and effectiveness of our proposed lightweight model that can be deployed on edge devices for real-world applications.