Enhanced Approaches to Similarity Analysis on Car Accident Reconstruction

Car accidents have significant repercussions for all parties involved, often leading to challenges in determining accountability. In the aftermath of such incidents, collaboration among vehicle forensic analysts, law enforcement officers, and insurance companies is crucial to gather substantial evidence for case resolution. To enhance the accuracy of information provision, this paper introduces the application of diverse mathematical techniques to assess similarities between meticulously reconstructed car accidents, including those involving autonomous vehicles, and corresponding real-world video footage. The primary aim of our study is to offer support to forensic investigators, insurance companies, academicians, and researchers in acquiring and leveraging precise information as evidential material in legal proceedings. Employing Car Learning to Act (CARLA), a simulation platform, we reconstruct car accident videos using samples from publicly available datasets. A comprehensive comparative analysis of these reconstructed videos are conducted, utilizing various mathematical algorithms, including Cosine similarity, Pearson Coefficient, and Euclidean distance, as the foundation for similarity assessments. Our simulation results show the optimal performance of the Pearson Coefficient and Euclidean distance coefficient in validating the accuracy of the reconstructed videos. In conclusion, our study contributes valuable insights to enhance the reliability of reconstructed accident scenarios, particularly in the context of autonomous vehicles.