CCTV 장면에서의 물피도주 차량 탐지 자동화 시스템
CCTV-based hit-and-run car detection automated system
1
조회
0
다운로드
1
인용
0
참고문헌
키워드
초록
사람들의 차량 보유 대수가 증가함에 따라 물피도주 사건도 증가하고 있지만, 수사 과정에서 장시간의 CCTV 영상을 수동으로 검수하므로 용의자를 식별하는 데 많은 시간이 필요하고 비효율적이다. 따라서, 불필요한 인력 소모를 줄이기 위해서는 물피도주 피의자 식별 자동화 시스템의 구축이 필수적이다. 본 논문에서는 CCTV 영상에 인공지능 기술을 활용해 CCTV 영상에서 '물피도주' 사건의 시점을 자동으로 탐지하고 가해 차량을 무엇인지 추적하는 시스템을 제안한다. 이 시스템은 CCTV 영상을 수신하면 피해 차량 선택, 사고 시점 추정, 가해 차량 추정의 과정을 거치며 가장 확률이 높은 물피도주 가해 차량을 탐지한다. 기존 차 사고 차량 사고 검출 방식의 취약점을 보완하고, 사고와 무관한 주변 객체로 인한 오류를 줄이기 위해 객체 분할과 객체 탐지 모델, 그리고 객체 추적과 깊이 추정 모델을 활용하여 주변 객체들을 별도로 분류하는 과정이 도입되었다. 또한 CCTV 영상에 촬영된 국내 차량의 탐지 성능을 향상하기 위해 별도로 구성한 데이터 세트를 통해 객체 탐지 성능을 향상했고, 실제 CCTV에서 촬영된 물피도주 사례들에 이 시스템을 적용했을 때 사고 발생 시점을 정확하게 검출하였고, HD 해상도의 영상에서 27fps 이상의 처리 속도를 보인다
Abstract
As the number of vehicles owned by individuals increases, so does the occurrence of hit-and-run incidents. However, the investigation process is time-consuming and inefficient due to the need for manual review of long-duration CCTV footage to identify suspects. Therefore, it is essential to establish an automated system for identifying hit-and-run suspects to reduce unnecessary manpower consumption. This paper proposes a system that utilizes artificial intelligence technology on CCTV footage to automatically detect the moment of a 'hit-and-run' incident and track the offending vehicle. Upon receiving CCTV footage, the system undergoes processes of selecting the victim vehicle, estimating the time of the accident, and estimating the offending vehicle, thereby detecting the most probable hit-and-run offending vehicle. To address the vulnerabilities of existing vehicle accident detection methods and to reduce errors caused by irrelevant surrounding objects, the process of object segmentation and detection models, as well as object tracking and depth estimation models, were introduced to separately classify surrounding objects. Additionally, to enhance the detection performance of domestic vehicles captured in CCTV footage, a separate dataset was compiled to improve object detection performance. When this system was applied to actual hit-and-run cases recorded by CCTV, it accurately detected the time of the accident and demonstrated a processing speed of over 27fps in HD resolution video.
참고문헌 (21)
- KRIZHEVSKY, Alex, "Imagenet classification with deep convolutional neural networks", Advances in neural information processing systems (2012)
- HE, Kaiming, "Deep residual learning for image recognition", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.770~778 (2016)
- TAN, Mingxing, "Efficientnet: Rethinking model scaling for convolutional neural networks", International conference on machine learning, pp.6105~6114 (2019), PMLR
- GIRSHICK, Ross, "Fast r-cnn", Proceedings of the IEEE international conference on computer vision, pp.1440~1448 (2015)
- REDMON, Joseph, "You only look once : Unified, real-time object detection", Proceedings of the IEEE conference on computer vision and pattern recognition, pp.779~788 (2016)
- HUANG, Huimin, "Unet 3+: A full-scale connected unet for medical image segmentation", ICASSP 2020-2020 IEEE international conference on acoustics, speech and signal processing (ICASSP), pp.1055~1059 (2020), IEEE
- CHEN, Liang-Chieh, "Deeplab: Semantic image segmentation with deep convolutional nets, atrous convolution, and fully connected crfs", IEEE transactions on pattern analysis and machine intelligence, 40(4), pp.834~848 (2017)
- WANG, Wenhai, "Internimage: Exploring large-scale vision foundation models with deformable convolutions", Proceedings of the IEEE, 2023, 14408~14419
- JEONG, Woo Jin, "Car accident detection in parking lots", 2014 International Conference on Electronics, Information and Communications (ICEIC), pp.1~2 (2014), IEEE
- REYNOLDS, Douglas A., "Encyclopedia of biometrics", pp.659~663 (2009)
- CHITALE, Pranav, "Smart Accident Recognition and Alerting System for Edge Devices", 2020 IEEE 5th International Conference on Computing Communication and Automation (ICCCA), pp.486~491 (2020), IEEE
- (2023)[URL]
- FARNEBÄCK, Gunnar, "Two-frame motion estimation based on polynomial expansion", Image Analysis: 13th Scandinavian Conference, SCIA 2003, 13, pp.363~370 (2003), Springer
- Silberman, Nathan, "Indoor segmentation and support inference from rgbd images", Computer Vision–ECCV 2012: 12th European Conference on Computer Vision, 12 (2012), Springer
- ZHANG, Yifu, "Bytetrack: Multi-object tracking by associating every detection box", European Conference on Computer Vision, pp.1~21 (2022), Springer
- LIN, Tsung-Yi, "Microsoft coco: Common objects in context", Computer Vision–ECCV 2014: 13th European Conference, 13, pp.740~755 (2014), Springer International Publishing