Educational project
Synthetic Aperture Radar (SAR) image simulation of parametric objects
SUPSI Image Focus
The thesis presents an integrated system composed of three modules for the simulation and validation of Synthetic Aperture Radar (SAR) images of industrial infrastructures.
The first module is a parametric oil tank generator that produces 3D models in both fixed-roof and floating-roof variants, ensuring geometries faithful to real-world structures.
The second module introduces a timeline- and keyframe-based animation system capable of sampling movements at the actual radar sensor acquisition times, accurately reproducing motion effects.
The third module implements a correlation between real and simulated images using 2D cross-correlation via FFT.
The system’s output is a synthetic dataset designed for training artificial intelligence models aimed at the automatic recognition of industrial objects.
The first module is a parametric oil tank generator that produces 3D models in both fixed-roof and floating-roof variants, ensuring geometries faithful to real-world structures.
The second module introduces a timeline- and keyframe-based animation system capable of sampling movements at the actual radar sensor acquisition times, accurately reproducing motion effects.
The third module implements a correlation between real and simulated images using 2D cross-correlation via FFT.
The system’s output is a synthetic dataset designed for training artificial intelligence models aimed at the automatic recognition of industrial objects.
The thesis presents a system for simulating and validating SAR images of industrial infrastructures, structured into three distinct but integrated modules.
The first component is a parametric generator of oil tanks, available in fixed- and floating-roof variants, with elements such as ladders, walkways, railings, and pipelines scaled and repositioned without distortion. The system produces clean and repeatable geometric models from input parameters, facilitating integration into the simulator.
The second component introduces an animation system with timeline and keyframe control, operating in both frame-based and time-based modes, and — most importantly — capable of sampling movements at the actual sensor acquisition times. This enables faithful reproduction of motion effects (such as blurring and aliasing) and supports batch simulations and time-series analyses.
The third component implements a correlation system between real and simulated images, based on 2D cross-correlation via FFT with an angular sweep around the estimated orientation. This approach retrieves translation and rotation parameters by maximizing similarity, allowing the correct rotation angle of the simulated object to be identified.
All modules were validated through geometric verification, analytical testing, and both quantitative and qualitative comparisons. Overall, the system reduces manual errors, increases experimental repeatability, and provides a solid foundation for future extensions in modeling, animation, and correlation.
The first component is a parametric generator of oil tanks, available in fixed- and floating-roof variants, with elements such as ladders, walkways, railings, and pipelines scaled and repositioned without distortion. The system produces clean and repeatable geometric models from input parameters, facilitating integration into the simulator.
The second component introduces an animation system with timeline and keyframe control, operating in both frame-based and time-based modes, and — most importantly — capable of sampling movements at the actual sensor acquisition times. This enables faithful reproduction of motion effects (such as blurring and aliasing) and supports batch simulations and time-series analyses.
The third component implements a correlation system between real and simulated images, based on 2D cross-correlation via FFT with an angular sweep around the estimated orientation. This approach retrieves translation and rotation parameters by maximizing similarity, allowing the correct rotation angle of the simulated object to be identified.
All modules were validated through geometric verification, analytical testing, and both quantitative and qualitative comparisons. Overall, the system reduces manual errors, increases experimental repeatability, and provides a solid foundation for future extensions in modeling, animation, and correlation.