Background. Automated surveillance systems using thermal images from UAVs struggle to detect and identify objects in real time, especially against countermeasures. Existing methods focus on motion segmentation rather than object identification and lack efficient localization mechanisms. Objective. The paper aims to develop and validate the Dynamic Thermal Portrait (DTP) algorithm – a lightweight two-factor method for rapid localisation of the target search zone within large thermal image frames, forming the first cascade stage before deep identification. The method is intended to shift from syntactic shape classification to physics-based analysis of micro-movements and thermal field dynamics. Methods. The DTP algorithm processes pairs of consecutive thermal frames (720×576 px) captured at ranges up to 4000 m with a small temporal shift (). A thermal mask () is constructed via Gaussian filtering and threshold segmentation with morphological dilation; a motion mask () is derived via absolute frame difference analysis with analogous post-processing. Pixel-by-pixel logical conjunction , followed by connected component analysis with area filtering, yields a bounding box that rapidly isolates the target zone within the full frame. Results. The proposed DTP method achieves approximately 185 FPS on CPU architectures – approaching ViBe speed and exceeding YOLOv8n by a factor of 4-5 – while operating on full-resolution frames without pre-cropping. The bounding-box output provides a compact region of interest for subsequent classification stages. Two-factor verification effectively filters static thermal decoys through micro-motion confirmation and suppresses dynamic background interference via the thermal pre-filter. Conclusions. The Dynamic Thermal Portrait method provides a computationally efficient first-stage detector that rapidly localises the target search zone in large thermal frames, enabling downstream identification systems to operate on a focused region rather than the full image. The physics-based two-factor verification distinguishes real targets from high-quality imitations, forming the methodological basis for a new generation of cascade intelligent monitoring systems under active countermoves and camouflage.
Background. Accurately estimating paved parking areas is vital for urban planning and environmental monitoring. However, spectral similarities between parking lot and…
Background. Resource allocation problems involving interdependent homogeneous entities arise in many practical settings, including competitive project funding…
Background. Financial time series are characterized by nonlinearity, non-stationarity, and a high level of noise, which significantly complicates their modeling and…
You may also start an advanced similarity search for this article.