Interest point detection for reconstruction in high granularity tracking detectors

This paper presents an investigation of the use of interest point detection algorithms from image processing applied to reconstruction of interactions in high granularity tracking detectors. Their purpose is to extract keypoints from the data as input to higher level reconstruction algorithms, replacing the role of human operators in event selection and reconstruction guidance. Simulations of nu(mu) + (40)Ar -> mu(-) + p events with a nu(mu) energy of 0.7GeV in a small liquid argon time projection chamber are used as a concrete example of a modern high granularity tracking detector. Data from the simulations are used to characterize the localization of interest points to physical features and the efficiency of finding interest points associated with the primary vertex and track ends is measured at the chosen beam energy. A high degree of localization is found, with 93% of detected interest points found within 5mm of a physical feature at the simulated energy of 0.7GeV. Working in two 2D projections, the primary vertex and both track ends are found in both projections in 85% of events at 0.7GeV. It is also shown that delta electrons can be detected.