Robust Dynamic Object Detection Algorithms for Continuous-Capture mWidar Imaging

The Microwave Imaging Detection, Analysis, and Ranging (mWidar) system is a novel sensing technology capable of continuous-capture, high-resolution imaging in a variety of challenging environments. However, its unique signal characteristics pose significant challenges for dynamic object detection and tracking algorithms. In this work, three detectors are presented and evaluated for their effectiveness on simulated mWidar images: a baseline numerical peak detector, which has been used in previous work, a signal processing-based constant false alarm rate (CFAR) detector, and a heuristic based time-dependent peak finding (TDPF) detector. Each detector is evaluated on simulated mWidar images with increasing complexity, ranging from single-object scenes to multi-object scenarios with occlusions and interference. The results demonstrate that both the CFAR and TDPF detectors achieve near-perfect detection rates only when a significant number of false positives are allowed. However, the TDPF detector overall outperforms the CFAR detector by maintaining a high detection rate with fewer false positives. These findings suggest that positioning-informed detection methods, such as the TDPF or tracking-assisted methods, may be more suitable for dynamic object detection in mWidar data compared to traditional signal processing techniques.