Radar Micro-Doppler Human Activity Recognition with CNN

A radar does not need a camera to see what a person is doing: the moving torso, arms and legs shift the frequency of the reflected signal (micro-Doppler). This project turns the raw beat signal of a real 5.8 GHz FMCW radar (University of Glasgow "Radar signatures of human activities") into micro-Doppler spectrograms - range FFT, MTI clutter filter, STFT - and recognises six activities: walking, sitting down, standing up, picking up an object, drinking and falling. The classical ECE baseline (38 hand-crafted Doppler features + SVM, and a Random Forest) is compared honestly with a CNN on 128 x 128 spectrogram images (the specification's network and an improved, position-aware CNN), using a split by participant and an external test on four new people. GNU Octave re-runs the whole radar chain on the original files, a point-scatterer human model simulates micro-Doppler from physics, and a web application recognises uploaded recordings with a fall alarm. A complete MATLAB version is included.

Category: Electronics and Communication Engineering (Radar Signal Processing, FMCW Radar, Micro-Doppler Signatures, Assisted Living and Fall Detection, Machine Learning, Deep Learning / CNN, GNU Octave, MATLAB, Python)

Electronics & Communication Engineering Project E05 from ProTechSC Academy, with graphical abstract, video demonstration and project guidance.