Low-Cost Concealed Weapon Detection for School Environments Using Acoustic Signatures

Over the past twenty years, concealed weapon detection (CWD) has been a great security concern. This concern has been compounded by recent events regarding compromised school security in the United States. In this paper, a pipeline, algorithm, and guidelines for CWD for school environments have been proposed. Acoustic signals were utilized for CWD because this signaling technology offers minimum health and privacy risks paired with maximum utility and cost-effectiveness. Acoustic signatures generated from the reflections of acoustic signals have been shown to vary based on target composition and shape. A library of acoustic signatures was compiled in order to identify concealed weapons. This was accomplished by projecting sound at frequencies 10 kilohertz through 16 kilohertz in 1 kilohertz increments at objects commonly found in school environments. A Fast Fourier Transform was then applied to the reflected waveform to extract a scalar feature from each acoustic signature. Then, a number of machine learning algorithms were used to create a mathematical model in order to predict weapon presence from acoustic input. Using Extremely Random Forest algorithm, weapons were detected with 88% accuracy.