Learning augmented standoff Concealed Weapon Detection
- Resource Type
- Conference
- Authors
- Malik, Muhammad Usman; Ahmed, Waqas; Majid, Abdul; Yaqoob, Zahid; Rafique, Abid
- Source
- 2017 14th International Bhurban Conference on Applied Sciences and Technology (IBCAST) Applied Sciences and Technology (IBCAST), 2017 14th International Bhurban Conference on. :830-837 Jan, 2017
- Subject
- Aerospace
Bioengineering
Communication, Networking and Broadcast Technologies
Computing and Processing
Engineered Materials, Dielectrics and Plasmas
Fields, Waves and Electromagnetics
Robotics and Control Systems
Signal Processing and Analysis
Weapons
Training
Radar detection
Signal analysis
Radar signal processing
Antennas
Concealed Weapon Detection
FMCW
Machine Learning
- Language
- ISSN
- 2151-1411
Polarimetric radar is one of the well known techniques for Concealed Weapon Detection. Usually a tri-static configuration is used, whereby a transmitter transmits energy with a specified polarization (horizontal or vertical) and two separate receivers receive the reflected energy. One of the receivers is co-polarized (CO) and the other is cross-polarized (X). The ratio of the CO and X signals will be high when the person is not carrying any weapon (no-threat) and that the ratio will be low in case of any concealed weapon (threat). One of the challenging tasks in such a system is to find the threshold for this ratio; a fixed threshold usually generates excessive false negatives and false positives depending on whether it is high or low enough. This paper presents a learning based approach and evaluates different classifiers to determine this decision. Actual data from 27 subjects demonstrate that J48 (decision-tree) based classifier can achieve an accuracy of 88% at standoff distances (∼18m), enabling the system to be used in practical scenarios.