The identification of minor impact collisions in a long video for detecting property damages caused by fleeing vehicles using three-dimensional convolutional neural network

Abstract

A parked vehicle dama ged by a hit-and-run can only be r e pair ed at the expense of the owner, unless the fleeing vehicle is identified and the dri v er appr ehended. Identifying the fleeing vehicle involves using a video investigation method that sear c hes for perpetrators thr ough CCTV foota ge of the crime scene. When the length of the r ecorded video is long, the inv estigation may r equir e an extended amount of time from the inv estigator, r esulting in an added burden on their daily w ork. Some commer cial companies are using object recognition and tr ac king tec hnology to detect hit-and-run incidents; however, detecting small movements of a vehicle during a minor collision still remains a c hallenge . Therefore , there is a need for a system that can detect small movement in a vehicle in a lengthy video . A utomatic recognition and tr ac king r equir e a sufficient amount of training dataset. However, such a dataset for hit-and-run incidents is not pub licl y av aila b le. One of the r easons behind this scarcity is that it may violate personal information pr otection acts. On the other hand, instead of using real accident videos, we could use actors to simulate such accident scenes. Although this may be feasib le, cr eating such a dataset would r equir e substantial costs. In this paper, we describe a new dataset for hit-and-run incidents. We collected 833 hit-and-run videos by r ecr eating a parking lot using miniaturized cars. This dataset has been made pub licl y av aila b le thr ough Ka ggle. We used thr ee-dimensional conv olution neur al netw ork, whic h is fr equentl y used in the field of action recognition, to detect small movements of vehicles during hit-and-run incidents. In addition, the proportion of the area that surrounds the target vehicle to the min-max box of the vehicle itself and the length of the input frame are varied to compare the accuracy. As a result, we wer e a b le to achiev e better accurac y by using the lo w est proportion and the shortest input fr ame . © The Author(s) 2024.