2-D Point Cloud Transformation Technique to Achieve Range–Angle Invariant Physical Activity Classification Using a Single mmWave Radar

This letter presents a 2-D point cloud data transformation technique to realize a range–angle invariant physical activities classification system using a single mmwave radar and deep convolutional neural network. The proposed 2-D point cloud transformer module applies normalization and translation techniques on point cloud data. This module compensates the range and angle variations in data due to position change of the user and helps to achieve user position invariant classifier or single angle classifier, which can accurately classify activities at multiple positions (having range and angular variations) based on training data from a single position. The proposed transformation technique eliminates the requirement of additional data collection and retraining the classifier. We present experimental results on a real-world dataset to validate the efficacy of the proposed technique.