Article Details

Abstract

This paper surveys the technologies for Anomalous Stationary Object Detection in highway scenarios, points out that traditional background modeling methods suffer from poor dynamic adaptability and lack the ability of object qualitative classification, and notes that pure Deep Learning solutions often face problems of insufficient real-time performance and frequent missed detections due to GPU resource constraints. This research proposes a three-stage framework that integrates combinational background modeling with Deep Learning. The first stage constructs dual-background models by adopting an improved Gaussian mixture model combined with a combinational caching mechanism to enhance recall. The second stage accurately localizes candidate regions through dual-background frame differencing and morphological operations so as to improve precision. The third stage accomplishes qualitative object classification via YOLOv8m. Experiments are conducted on a dataset consisting of 1 700 videos covering seven categories of litter objects, and the results demonstrate that the model achieves a precision of 95.3% and a recall of 82.7%. The model is currently successfully deployed on SE9 edge devices and applied in a provincial smart highway system.

Keywords

combinational background modeling; Anomalous Stationary Object Detection; Deep Learning; edge computing; highway

References

1. [1] Shi Yonglu, Li Peiyun, Xie Zhiyu, etc Method and system for detecting anomalous stationary targets on highways based on combined backgrounds: 202411649378.4 [P]. 20225-03-11
2. [2] Ministry of Transport. Statistical Bulletin on the Development of the Transportation Industry in 2024 [R/OL]. (June 12, 2025) https://xxgk.mot.gov.cn/2020/jigou/zhghs/202506/t20250610_4170228.html.
3. [3] REYNOLDS D. Gaussian Mixture Models [M].New York：Springer，2009：659–663.
4. [4] Wang Yang, Zhou Jiaogen, Yan Jun, etc A review of anomaly detection methods for surveillance videos based on deep learning [J]. Chinese Journal of Image and Graphics, 2025, 30 (3): 615-640
5. [5] ZIVKOVIC Z，HEIJDEN F V D.Efficient Adaptive Density Estimation per Image Pixel for the Task of Background Subtraction [J].Pattern Recognition Letters，2006，27（7）：773-780.
6. [6] LIU W，LUO W，LIAN D，et al. Future Frame Prediction for Anomaly Detection-A New Baseline [C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition.Salt Lake City：IEEE，2017：6536-6545.
7. [7] BRAHAM M，DROOGENBROECK M V.Deep Background Subtraction with Scene-specific Convolutional Neural Networks [C]//2016 International Conference on Systems, Signals and Image Processing (IWSSIP).Bratislava：IEEE，2016：1-4.
8. [8] STAUFFER C，CHRIS，GRIMSON W E L. Adaptive Background Mixture Models for Real-time Tracking [C]//1999 IEEE Computer Society Conference on Computer Vision and Pattern Recognition. Fort Collins：IEEE，1999：246-252.
9. [9] XIA Y，OUYANG C Y. Dynamic Image Background Modeling Method for Detecting Abandoned Objects in Highway [J].Journal of Zhejiang University ：Engineering Science，2020，54（7）：1249-1255.
10. [10] Xu Guoliang, Zhou Hang, Yuan Liangyou Human Ghost Suppression Algorithm Using Gaussian Mixture and Topological Structure [J]. Journal of Intelligent Systems, 2021, 16 (2): 294-302
11. [11] Huang Yu, Yin Changming, Zhou Shuren Adaptive Gaussian Mixture Model Background Modeling Based on YCbCr [J]. Computer Engineering and Science, 2015, 37 (1): 152-156
12. [12] ZIVKOVIC Z. Improved Adaptive Gaussian Mixture Model for Background Subtraction [C]//Proceedings of the 17th International Conference on Pattern Recognition.Cambridge：IEEE，2004：28-31.
13. [13] REDMON J，FARHADI A. YOLOv3: An Incremental Improvement [J/OL].arXiv:1804.02767 [cs.CV].(2018-04-08).https://arxiv.org/abs/1804.02767.
14. [14] YASEEN M. What is YOLOv8: An In-depth Exploration of the Internal Features of the Next-generation Object Detector [J/OL].arXiv:2408.15857 [cs.CV].(2024-08-28).https://arxiv.org/abs/2408.15857.
15. [15] Wang Xin, Tian Tian, Tian Jinwen VideoSAR Moving Target Shadow Detection Method Based on Background Modeling [J]. Journal of Intelligent Systems, 2022, 17 (1): 59-68
16. [16] BARNICH O，VAN DROOGENBROECK M. ViBe: A Universal Background Subtraction Algorithm for Video Sequences [J].IEEE Transactions on Image Processing，2011，20（6）：1709-1724.

About the authors

Shi Yonglu (1988-), male, Han ethnicity, from Bazhong, Sichuan, engineer, bachelor's degree, research interests include machine vision, multimodality and computer science and technology; Yang Yong (1990-), male, Han ethnicity, from Ziyang, Sichuan, master's degree, research interests include software engineering; Han Kewei (1986-), male, Han ethnicity, from Xianyang, Shaanxi, master's degree, research interests include multimodal large models; Li Peiyun (1992-), female, Han ethnicity, from Guangyuan, Sichuan, bachelor's degree, research interests include industry applications of large models; Zhang Xiao (1988-), Han ethnicity, from Guang'an, Sichuan, master's degree, research interests include AI product design; Li Jiahui (1995-), female, Han ethnicity, from Leshan, Sichuan, master's degree, research interests include object detection.