Construction of VSSM-CNN detection network for nighttime road traffic accidents
Article Text (Baidu Translation)
-
摘要: 为提升夜间场景下道路交通事故的自动检测性能，基于视觉状态空间模型(VSSM)与卷积神经网络(CNN)构建了VSSM-CNN编码器-解码器架构，提出了一种面向该场景的无监督交通事故检测框架；参考特征融合思路，并基于已有的可见光图像作为外观特征，进一步采用循环全对场变换(RAFT)光流估计算法结合卷积长短期记忆网络(ConvLSTM)模块提取并处理视频序列的细粒度光流信息，以表征交通运动状态；将外观与运动两类特征融合后输入至以视觉状态空间模型为主干网络的特征编码器进行全局特征提取，并采用卷积神经网络作为解码器架构，逐层恢复图像以强化局部细节，从而加强特征提取效率与利用效果；采取均方误差、平均绝对误差与结构相似性三重损失函数组合，通过贝叶斯优化确定权重比例，以提升模型在夜间图像重构时对异常结构与噪声的鲁棒性。研究结果表明：所提方法的受试者工作特征曲线下面积(ROC-AUC)与精确率-召回率曲线下面积(PR-AUC)分别为0.818与0.765，较传统生成式网络分别提升了22.6%和20.3%，其中ROC-AUC在与多种现有方法的比较中取得了最高值；三重损失函数在消融试验中展现出了最强的模型性能提升能力；研究方法取得了最低的模型复杂度与较优的检测速度，充分展现其异常识别的稳定性与部署应用的潜力。研究成果有效提升了夜间场景下的交通事故检测能力，可进一步推广至低光照、低能见度场景下的交通事故检测，并为相关的应用提供可行的技术方案与参考。Abstract: To enhance the automatic detection performance of road traffic accidents in nighttime scenarios, a VSSM-CNN encoder-decoder architecture was constructed based on a visual state space model (VSSM) and a convolutional neural network (CNN), and an unsupervised traffic accident detection framework oriented to this scenario was proposed. By referencing the idea of feature fusion and based on existing visible-light images as appearance features, fine-grained optical flow information of video sequences was further extracted and processed using a recurrent all-pairs field transform (RAFT) optical flow estimation algorithm combined with a convolutional long short-term memory (ConvLSTM) module to represent traffic motion states. The two types of features, appearance and motion, were fused and input into a feature encoder with the VSSM as a backbone network for global feature extraction, and a CNN was adopted as a decoder architecture to recover images layer by layer to enhance local details, so as to strengthen feature extraction efficiency and utilization effects. A triple loss function combination of mean squared error, mean absolute error, and structural similarity was adopted, and weight proportions were determined through Bayesian optimization to improve the robustness of the model to abnormal structures and noise during nighttime image reconstruction. Research results indicate that the area under the receiver operating characteristic curve (ROC-AUC) and area under the precision-recall curve (PR-AUC) of the proposed method are 0.818 and 0.765, respectively, which improve by 22.6% and 20.3%, respectively, compared with traditional generative networks; among them, the ROC-AUC achieves the highest value in the comparison with multiple existing methods; the triple loss function shows the strongest model performance improvement ability in ablation experiments; the research method achieves the lowest model complexity and a favorable detection speed, fully demonstrating its stability in anomaly recognition and potential for deployment and application. The research results effectively improve the traffic accident detection ability in nighttime scenarios, can be further extended to traffic accident detection in low-illumination and low-visibility scenarios, and provide a feasible technical solution and reference for related applications.
-
图 1 夜间环境交通事故检测网络
Figure 1. Traffic accident detection network in nighttime environments
图 4 不同光流提取方法在自采集数据上的提取效果对比
Figure 4. Comparison of extraction effects of different optical flow estimation methods on self-collected data
图 6 夜间交通事故样本及其重构误差分布热力图
Figure 6. Accident samples under nighttime conditions and their heatmaps of reconstruction error distributions
表 1 数据集信息
Table 1. Dataset information
数据集 片段数 帧数 非事故(公开数据集HWID12) 1 074 26 850 非事故(自采集数据集) 600 15 000 事故(公开数据集CADP+网络视频) 342 8 550 
下载: 导出CSV
表 2 消融试验结果
Table 2. Results of the ablation study
模型 运动信息 VSSM 损失函数 ConvLSTM_module ROC-AUC PR-AUC 基础模型 × × × × 0.592 0.562 拓展模型1 √ √ × √ 0.730 0.695 拓展模型2 √ × √ √ 0.759 0.769 拓展模型3 × √ √ √ 0.806 0.750 拓展模型4 √ √ √ × 0.808 0.752 本研究模型 √ √ √ √ 0.818 0.765
下载: 导出CSV
表 3 各模型性能对比
Table 3. Comparison of the performance of various models
模型 召回率 F1分数 F2分数 ROC-AUC PR-AUC 运算量/GFLOPs 推理速度/(样本·s-1) C3D 0.546 0.636 0.578 0.757 0.722 468.43 9.18 Conv-AE 0.520 0.670 0.571 0.702 0.713 136.72 7.87 3DConv-AE 0.894 0.494 0.675 0.591 0.505 102.96 8.54 MemAE 0.574 0.715 0.623 0.778 0.771 635.83 7.79 MNAD 0.543 0.668 0.587 0.713 0.711 188.12 7.04 MONAD 0.554 0.704 0.606 0.809 0.782 1 271.66 6.82 本研究模型 0.593 0.668 0.621 0.818 0.765 5.19 7.45
下载: 导出CSV
-
[1] 王晨, 周威, 严隽逸, 等. 一种用于道路交通事故自动检测的改进双流网络[J]. 中国公路学报, 2023, 36(5): 185-196.WANG Chen, ZHOU Wei, YAN Jun-yi, et al. Improved two-stream network for vision-based traffic accident detection[J]. China Journal of Highway and Transport, 2023, 36(5): 185-196. [2] 杨洋, 王文慧, 吴先宇, 等. 高速公路非常规交通事故研究综述[J]. 应用基础与工程科学学报, 2024, 32(3): 601-626.YANG Yang, WANG Wen-hui, WU Xian-yu, et al. Review of the research toward freeway unconventional traffic accidents[J]. Journal of Basic Science and Engineering, 2024, 32(3): 601-626. [3] WANG W C, YANG Y, YANG X B, et al. A negative binomial Lindley approach considering spatiotemporal effects for modeling traffic crash frequency with excess zeros[J]. Accident Analysis & Prevention, 2024, 207: 107741. [4] 黄思德, 黄荣军, 张岩坤, 等. 基于监控视频的交通事故自动识别算法[J]. 公路交通科技, 2024, 41(1): 169-176.HUANG Si-de, HUANG Rong-jun, ZHANG Yan-kun, et al. An algorithm for automatic traffic incident detection based on surveillance video[J]. Journal of Highway and Transportation Research and Development, 2024, 41(1): 169-176. [5] 郭延永, 刘佩, 袁泉, 等. 网联自动驾驶车辆道路交通安全研究综述[J]. 交通运输工程学报, 2023, 23(5): 19-38. doi: 10.19818/j.cnki.1671-1637.2023.05.002 GUO Yan-yong, LIU Pei, YUAN Quan, et al. Review on research of road traffic safety of connected and automated vehicles[J]. Journal of Traffic and Transportation Engineering, 2023, 23(5): 19-38. doi: 10.19818/j.cnki.1671-1637.2023.05.002 [6] 公安部办公厅, 国家卫生健康委办公厅. 关于健全完善道路交通事故警医联动救援救治长效机制的通知: 公交管〔2020〕161号[EB/OL]. (2020-07-02)[2025-08-04]. http://www.nhc.gov.cn/vzvgj/s3594q/202007/c41e38d7db744ad9890ecb5d710ab59b.shtml .General Office of the Ministry of Public Security, General Office of the National Health Commission. Notice on improving the long-term mechanism of police-medical joint rescue and treatment for road traffic accidents: Gong Jiao Guan 〔2020〕 No. 161[EB/OL]. (2020-07-02)[2025-08-04].http://www.nhc.gov.cn/vzvgj/s3594q/202007/c41e38d7db744ad9890ecb5d710ab59b.shtml .[7] 阎莹, 王玉莹, 周旋, 等. 基于混合模型的自动驾驶车辆事故严重程度影响因素分析[J]. 交通运输工程学报, 2025, 25(1): 184-196. doi: 10.19818/j.cnki.1671-1637.2025.01.013 YAN Ying, WANG Yu-ying, ZHOU Xuan, et al. Analysis of autonomous vehicle accident severity factors based on hybrid model[J]. Journal of Traffic and Transportation Engineering, 2025, 25(1): 184-196. doi: 10.19818/j.cnki.1671-1637.2025.01.013 [8] 杨洋, 贺昆, 王云鹏, 等. 面向动态交通流的高速公路事故风险模型空间移植研究[J]. 交通运输系统工程与信息, 2023, 23(3): 174-186.YANG Yang, HE Kun, WANG Yun-peng, et al. Spatial transplantation for modeling of freeway traffic crash risk based on dynamic traffic flow[J]. Journal of Transportation Systems Engineering and Information Technology, 2023, 23(3): 174-186. [9] 牛世峰, 母俊杰, 濮泽昱, 等. 基于错时序因果变量和分目标训练的车辆事故风险分类分层评估方法[J]. 交通运输工程学报, 2025, 25(6): 293-306. doi: 10.19818/j.cnki.1671-1637.2025.06.024 NIU Shi-feng, MU Jun-jie, PU Ze-yu, et al. Hierarchical assessment method for vehicle accident risk classification based on temporal misalignment causal variables and sub-objective training[J]. Journal of Traffic and Transportation Engineering, 2025, 25(6): 293-306. doi: 10.19818/j.cnki.1671-1637.2025.06.024 [10] 杨洋, 袁振洲, 王印海, 等. 基于WOMDI-Apriori算法的高速公路交通事故风险识别[J]. 交通工程, 2021, 21(6): 1-10, 16.YANG Yang, YUAN Zhen-zhou, WANG Yin-hai, et al. Freeway crash risk identification based on a new improved method of WOMDI-Apriori algorithm[J]. Journal of Transportation Engineering, 2021, 21(6): 1-10, 16. [11] BATANINA E, BEKKOUCH I E I, YOUSSRY Y, et al. Domain adaptation for car accident detection in videos[C]//IEEE. 2019 Ninth International Conference on Image Processing Theory, Tools and Applications (IPTA). New York: IEEE, 2019: 1-6. [12] FANG J W, QIAO J H, BAI J, et al. Traffic accident detection via self-supervised consistency learning in driving scenarios[J]. IEEE Transactions on Intelligent Transportation Systems, 2022, 23(7): 9601-9614. doi: 10.1109/TITS.2022.3157254 [13] PASHAEI A, GHATEE M, SAJEDI H. Convolution neural network joint with mixture of extreme learning machines for feature extraction and classification of accident images[J]. Journal of Real-time Image Processing, 2020, 17(4): 1051-1066. doi: 10.1007/s11554-019-00852-3 [14] IJJINA E P, CHAND D, GUPTA S, et al. Computer vision-based accident detection in traffic surveillance[C]//IEEE. 2019 10th International Conference on Computing, Communication and Networking Technologies (ICCCNT). New York: IEEE, 2019: 1-6. [15] JIANG Y, WANG Y H, ZHAO M H, et al. Nighttime traffic object detection via adaptively integrating event and frame domains[J]. Fundamental Research, 2025, 5(4): 1633-1644. doi: 10.1016/j.fmre.2023.08.004 [16] YUE Y G, ZHANG S, WU Z Y, et al. Research on nighttime road visibility monitoring based on video images[J]. Traffic Injury Prevention, 2026, 27(3): 287-295. doi: 10.1080/15389588.2025.2495203 [17] HOSSAIN A, SUN X D, SHAHRIER M, et al. Exploring nighttime pedestrian crash patterns at intersection and segments: Findings from the machine learning algorithm[J]. Journal of Safety Research, 2023, 87: 382-394. doi: 10.1016/j.jsr.2023.08.010 [18] REN J S, WANG W, WANG J W, et al. An unsupervised feature learning approach to improve automatic incident detection[C]//IEEE. 2012 15th International IEEE Conference on Intelligent Transportation Systems. New York: IEEE, 2012: 172-177. [19] YAO Y, XU M Z, WANG Y C, et al. Unsupervised traffic accident detection in first-person videos[C]//IEEE. 2019 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS). New York: IEEE, 2019: 273-280. [20] SINGH D, MOHAN C K. Deep spatio-temporal representation for detection of road accidents using stacked autoencoder[J]. IEEE Transactions on Intelligent Transportation Systems, 2019, 20(3): 879-887. doi: 10.1109/TITS.2018.2835308 [21] ZHOU W, WEN L H, ZHAN Y F, et al. An appearance-motion network for vision-based crash detection: Improving the accuracy in congested traffic[J]. IEEE Transactions on Intelligent Transportation Systems, 2023, 24(12): 13742-13755. doi: 10.1109/TITS.2023.3297589 [22] CHEN P, ZHANG W W, XIAO Z Y, et al. Traffic accident detection based on deformable frustum proposal and adaptive space segmentation[J]. Computer Modeling in Engineering & Sciences, 2022, 130(1): 97-109. [23] 张奇, 周威, 胡伟超, 等. 融合递进式域适应和交叉注意力的事故检测方法[J]. 计算机工程与应用, 2025, 61(6): 349-360.ZHANG Qi, ZHOU Wei, HU Wei-chao, et al. Accident detection method integrating progressive domain adaptation and cross-attention[J]. Computer Engineering and Applications, 2025, 61(6): 349-360. [24] TRAN T M, BUI D C, NGUYEN T V, et al. Transformer-based spatio-temporal unsupervised traffic anomaly detection in aerial videos[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2024, 34(9): 8292-8309. doi: 10.1109/TCSVT.2024.3376399 [25] LIANG R Q, LI Y M, YI Y X, et al. A memory-augmented multi-task collaborative framework for unsupervised traffic anomaly detection in driving videos[J]. Pattern Recognition, 2025, 168: 111789. doi: 10.1016/j.patcog.2025.111789 [26] LUO W X, LIU W, LIAN D Z, et al. Video anomaly detection with sparse coding inspired deep neural networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2021, 43(3): 1070-1084. doi: 10.1109/TPAMI.2019.2944377 [27] ZHANG Y, NIE X S, HE R D, et al. Normality learning in multispace for video anomaly detection[J]. IEEE Transactions on Circuits and Systems for Video Technology, 2021, 31(9): 3694-3706. doi: 10.1109/TCSVT.2020.3039798 [28] QIE K, WANG J Y, LI Z H, et al. Recognition of occluded pedestrians from the driver's perspective for extending sight distance and ensuring driving safety at signal-free intersections[J]. Digital Transportation and Safety, 2024, 3(2): 65-74. doi: 10.48130/dts-0024-0007 [29] 石洋宇, 谢承杰, 郑棣文, 等. 基于Mamba-CNN的多尺度异常行为检测方法[J/OL]. 北京航空航天大学学报, (2024-11-15). https://doi.org/10.13700/j.bh.1001-5965.2024.0416 .SHI Yang-yu, XIE Cheng-jie, ZHENG Di-wen, et al. Multi-scale anomaly behavior detection method based on Mamba-CNN[J/OL]. Journal of Beijing University of Aeronautics and Astronautics, (2024-11-15).https://doi.org/10.13700/j.bh.1001-5965.2024.0416 .[30] PENG Z L, GUO Z H, HUANG W, et al. Conformer: Local features coupling global representations for recognition and detection[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2023, 45(8): 9454-9468. doi: 10.1109/TPAMI.2023.3243048 [31] SRINIVAS A, LIN T Y, PARMAR N, et al. Bottleneck transformers for visual recognition[C]//IEEE. 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2021: 16514-16524. [32] TEED Z, DENG J. RAFT: Recurrent all-pairs field transforms for optical flow[C]//ECCV. Computer Vision-ECCV 2020. Berlin: Springer International Publishing, 2020: 402-419. [33] JIAO J B, LIU Y, LIU Y F, et al. VMamba: Visual state space model[C]//NeurIPS. Advances in Neural Information Processing Systems 37. Vancouver: Curran Neural Information Processing Systems Foundation, Inc. (NeurIPS), 2024: 103031-103063. [34] DOSOVITSKIY A, FISCHER P, ILG E, et al. FlowNet: Learning optical flow with convolutional networks[C]//IEEE. 2015 IEEE International Conference on Computer Vision (ICCV). New York: IEEE, 2015: 2758-2766. [35] SUN D Q, YANG X D, LIU M Y, et al. PWC-net: CNNs for optical flow using pyramid, warping, and cost volume[C]// IEEE. 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2018: 8934-8943. [36] CHEN H J, WANG Z, QIN H D, et al. CFDHI-net: Correlation-driven feature decoupling and hierarchical integration network for RGB-thermal semantic segmentation[J]. IEEE Transactions on Intelligent Transportation Systems, 2025, 26(10): 17173-17184. doi: 10.1109/TITS.2025.3581609 [37] FAN J, CHEN M, GU Z Y, et al. SSIM over MSE: A new perspective for video anomaly detection[J]. Neural Networks, 2025, 185: 107115. doi: 10.1016/j.neunet.2024.107115 [38] ISOLA P, ZHU J Y, ZHOU T H, et al. Image-to-image translation with conditional adversarial networks[C]//IEEE. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2017: 1125-1134. [39] KLEIN A, FALKNER S, BARTELS S, et al. Fast Bayesian optimization of machine learning hyperparameters on large datasets[C]//PMLR. Proceedings of the 20th International Conference on Artificial Intelligence and Statistics. Fort Lauderdale: PMLR, 2017: 528-536. [40] WU J, TOSCANO-PALMERIN S, FRAZIER P I, et al. Practical multi-fidelity Bayesian optimization for hyperparameter tuning[C]//PMLR. Proceedings of the 35th Uncertainty in Artificial Intelligence Conference. Fort Lauderdale: PMLR, 2020: 788-798. [41] SHAH A P, LAMARE J B, NGUYEN-ANH T, et al. CADP: A novel dataset for CCTV traffic camera based accident analysis[C]//IEEE. 2018 15th IEEE International Conference on Advanced Video and Signal Based Surveillance (AVSS). New York: IEEE, 2018: 1-9. [42] HUANG X H, HE P, RANGARAJAN A, et al. Intelligent intersection: Two-stream convolutional networks for real-time near-accident detection in traffic video[J]. ACM Transactions on Spatial Algorithms and Systems, 2020, 6(2): 1-28. [43] HASAN M, CHOI J, NEUMANN J, et al. Learning temporal regularity in video sequences[C]//IEEE. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2016: 733-742. [44] TRAN D, BOURDEV L, FERGUS R, et al. Learning spatiotemporal features with 3D convolutional networks[C]// IEEE. 2015 IEEE International Conference on Computer Vision (ICCV). New York: IEEE, 2015: 4489-4497. [45] GONG D, LIU L Q, LE V, et al. Memorizing normality to detect anomaly: Memory-augmented deep autoencoder for unsupervised anomaly detection[C]//IEEE. 2019 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE, 2019: 1705-1714. [46] PARK H, NOH J, HAM B. Learning memory-guided normality for anomaly detection[C]//IEEE. 2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE, 2020: 14372-14381. [47] DOSHI K, YILMAZ Y. Online anomaly detection in surveillance videos with asymptotic bounds on false alarm rate[C]//Springer International Publishing. Pattern Recognition-ICPR International Workshops and Challenges. Berlin: Springer International Publishing, 2021: 292-304. -
图(7) / 表(3)
计量
- 文章访问数: 16
- HTML全文浏览量: 12
- PDF下载量: 7
- 被引次数: 0
