Development of a Cost-Efficient Multi-Target Classification System Based on FMCW Radar for Security Gate Monitoring

Radar systems have a long history. Like many other great inventions, the origin of radar systems lies in warfare. Only in the last decade, radar systems have found widespread civil use in industrial measurement scenarios and automotive safety applications. Due to their resilience against harsh environments, they are used instead of or in addition to optical or ultrasonic systems. Radar sensors hold excellent capabilities to estimate distance and motion accurately, penetrate non-metallic objects, and remain unaffected by weather conditions. These capabilities make these devices extremely flexible in their applications. Electromagnetic waves centered at frequencies around 24 GHz offer high precision target measurements, compact antenna, and circuitry design, and lower atmospheric absorption than higher frequency-based systems. This thesis studies non-cooperative automatic radar multi-target detection and classification. A prototype of a radar system with a new microwave-radar-based technique for short-range detection and classification of multiple human and vehicle targets passing through a road gate is presented. It allows identifying different types of targets, i.e., pedestrians, motorcycles, cars, and trucks. The developed system is based on a low-cost 24 GHz off-the-shelf FMCW radar, combined with an embedded Raspberry Pi PC for data acquisition and transmission to a remote processing PC, which takes care of detection and classification. This approach, which can find applications in both security and infrastructure surveillance, relies upon the processing of the scattered-field data acquired by the radar. The developed method is based on an ad-hoc processing chain to accomplish the automatic target recognition task, which consists of blocks performing clutter and leakage removal with a frame subtraction technique, clustering with a DBSCAN approach, tracking algorithm based on the α-β filter to follow the targets during traversal, features extraction, and finally classification of targets with a classification scheme based on support vector machines. The approach is validated in real experimental scenarios, showing its capabilities incorrectly detecting multiple targets belonging to different classes (i.e., pedestrians, cars, motorcycles, and trucks). The approach has been validated with experimental data acquired in different scenarios, showing good identification capabilities.