Exploration of adaptive Hough algorithm variants for charged particle tracks finding for the ATLAS experiment for the LHC Run 4

The thesis presents an in-depth performance study of the Adaptive Hough transform (AHT) for particle tracking. In the classical Hough transform particle parameters are determined based on pixelized parameters space, which previously has been filled with lines corresponding to measurement point. Conversely, the AHT relies on unconstrained allowing any precision of parameters estimate. The AHT is a recursive algorithm and therefore uses a modest amount of memory resources even for high-precision estimates. In practice, it has been implemented using the stack of the parameter space sections to make it compatible with accelerator devices like GPU. Such an approach to particle track candidates finding results in high algorithm flexibility. The main objective of the thesis was to evaluate the algorithm tracking performance for the high pileup events (roughly 200 proton-proton interactions for each bunch crossing) that will be gathered by the ATLAS detector at High Luminosity LHC. In the algorithm testing stage, both single muon and pile-up events have been used. Simple AHT resulted in 100% efficiency for a single muon sample and very good tracking parameters estimation precision. However, a significant number of spurious Hough solutions have been observed. To reduce the number of obtained solutions, while preserving the high tracking efficiency three optimization methods have been implemented 1) counting the number of lines intersecting within the section, 2) evaluating the mean and standard deviation for the particles crossing points 3) data partitioning into wedges. The two first steps resulted in the number of solutions for single muon truth particles dropping from roughly 10 to 1.8. The optimization step requiring data division into wedges is relevant only for the pile-up dataset. Including all the optimization steps resulted in a significant decrease in the number of solutions for single truth particles in the pileup sample, from approximately 185 to 14.1 The efficiency at the same time was reduced from 99.1% to 93.2%.