Computer Vision for Crime Recognition Based on Skeleton Trajectories

Given the substantial amount of data generated daily by surveillance systems in urban areas, there is a growing necessity for automation in the crime detection process. Considering the limitations of the current approaches to detecting crime in surveillance videos, there is a need for a new approach that helps reduce human labor and its decision-making ability to ensure the safety of the public. The objective of this research to evaluate the accuracy of skeleton-based action recognition models within the crime domain and use the HR-Crime dataset as the reference point for comparison with other modalities.