CCTV 환경에서의 물피도주 사고 시점 추정을 위한 광학 흐름 및 객체 탐지 기반 자동화 시스템 = Automated System for Estimating Hit-and-Run Accident Time in CCTV Environments Based on Optical Flow and Object Detection

다국어 초록 (Multilingual Abstract)

CCTV footage." Hit-and-run incidents involve collisions with parked vehicles followed by fleeing the scene, which has been increasing annually. However, the current investigation method includes inefficient processes, such as investigators having to analyze CCTV footage manually for long periods. The improved system addresses key weaknesses of the existing system, such as video processing speed, false detection caused by surrounding objects, and complex algorithm structures. In particular, the optical flow calculation algorithm was optimized, and vectorized operations were introduced to improve processing efficiency. Additionally, the use of frame interval sampling techniques shortened analysis time. A major improvement involved completely removing the complex depth estimation model previously used in the system, and adopting a simplified approach that more accurately identifies the accident point through the analysis of object positions and movements on a 2D plane. As a result, computational complexity was significantly reduced, and the algorithm's efficiency was enhanced. Experimental results show that the improved system provides more efficient processing compared to the previous system, with a 67% reduction in false detections in nighttime footage. Moreover, it accurately detects the accident points across all test videos, and the improvements made to the system are expected to contribute to enhancing the efficiency of investigations into hit-and-run incidents.