摘要
—Chemical laboratory accidents such as spills, falls, and chemical burns occur even among well trained staff. Timely detection is therefore critical, yet existing monitoring solutions often involve tradeoffs between privacy, usability, and regulatory compliance. This paper proposes a privacy preserving method for detecting dangerous activities using a 24 GHz FMCW radar. Four participants performed four benign and two dangerous classes of motions in an empty room and in a fully stocked, operational chemistry laboratory. A two stage classifier is designed on Doppler spectrograms. First, the temporal evolution of the Doppler signatures is used to distinguish periodic from non periodic motions. In our datasets, periodic motions correspond to routine laboratory behaviors and are therefore treated as benign classes. Second, a peak velocity V peak is extracted from the spectrogram, and a single threshold derived from the empty room data is applied unchanged to both environments. The results show that the proposed approach separates benign and dangerous activities reliably in terms of peak velocity and achieves high accuracy, precision, and recall in both environments while requiring negligible computing resources. These findings suggest that low cost FMCW radar can provide a practical and privacy preserving option for continuous laboratory safety monitoring.