1.School of Traffic and Transportation, Beijing Jiaotong University, Beijing 100044, China
2.Key Laboratory of Railway Industry for Operational Active Safety Assurance and Risk Prevention and Control, Beijing Jiaotong University, Beijing 100044, China
3.State Key Laboratory of Advanced Rail Autonomous Operation, Beijing Jiaotong University, Beijing 100044, China
4.Beijing Engineering Research Center of Intelligent Perception and Service for Urban Traffic Information, Beijing Jiaotong University, Beijing 100044, China
5.College of Computer Science, Nankai University, Tianjin 300071, China
To solve the performance limitations of traditional object detection and tracking methods in complex scenes such as large scale variations and severe dense occlusion, an estimation method of passenger flow density based on CNN-Transformer fusion is proposed. Firstly, the VGG-19 backbone network is used to extract multi-scale features, and a Multi-scale Feature Fusion Module (MFFM) is constructed to effectively capture local details and spatial information by integrating feature maps of different resolutions. Secondly, a Multi-scale Perception Enhancement Module (MPEM) is introduced, combined with adaptive convolution and dilated convolution for lightweight design, to enhance the model's perception ability for complex contexts and occluded areas. Finally, a global dependency relationship is established through Transformer encoder to compensate for feature loss and further improve the robustness of the model. The results show that the proposed method performs well on the self-built Metro-platform dataset, the mean absolute error and mean square error are reduced to 18.4 and 23.3 respectively, and the reasoning time is 29.5 ms. Tests results on public crowd datasets such as ShanghaiTech-A, UCF-QNRF, and JHU-Crowd++ further validate generalization ability of the model. This method can provide a highly accurate and efficient solution for estimating passenger flow density in urban rail transit stations, and provide technical support for passenger safety management and operation optimization.
China Association of Metros. Interpretation of Statistics and Analysis Report of Urban Rail Transit in 2023 [J]. China Metros, 2024 (4): 15-17. in Chinese
[3]
ZHANGY Y, ZHOUD S, CHENS Q, et al. Single-Image Crowd Counting via Multi-Column Convolutional Neural Network [C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2016: 589-597.
[4]
ZHANGA, SHENJ, XIAOZ, et al. Relational Attention Network for Crowd Counting [C]// 2019 IEEE/CVF International Conference on Computer Vision (ICCV). New York: IEEE Press, 2019: 6787-6796.
WANGT, ZHANGT, ZHANGK, et al. Context Attention Fusion Network for Crowd Counting [J]. Knowledge-Based Systems, 2023, 271:110541.
[7]
QIANX, YUX, FA C. The Passenger Flow Counting Research of Subway Video Based on Image Processing [C]// 2017 29th Chinese Control and Decision Conference (CCDC). New York: IEEE Press, 2017: 5195-5198.
[8]
SHENGZ, TIANK, TIANQ, et al. A Faster R-CNN Based High-Normalization Sample Calibration Method for Dense Subway Passenger Flow Detection [C]// 2018 11th International Congress on Image and Signal Processing, BioMedical Engineering and Informatics (CISP-BMEI). New York: IEEE Press, 2018: 1-5.
[9]
HUX, ZHENGH, WANGW, et al. A Novel Approach for Crowd Video Monitoring of Subway Platforms [J]. Optik, 124 (22): 5301-5306.
WUJunyan, LIUXia, LIYazhuo. Application Research on Passenger Flow Real-Time Detection of Subway Station Platform Based on Deep Learning [J]. Journal of Jianghan University (Natural Science Edition), 2023, 51 (5): 67-74. in Chinese
LIRui, LIPing, DAIMingrui, et al. Crowd Counting Estimation Model Based on Multi-View Projection Fusion of Railway Passenger Station Video [J]. China Railway Science, 2022, 43 (3): 182-192. in Chinese
[14]
VASWANIA, SHAZEERN, PARMARN, et al. Attention Is All You Need [J]. ArXiv e-Prints, 2017: arXiv:
[15]
LIANGD K, CHENX W, XUW, et al. TransCrowd: Weakly-Supervised Crowd Counting with Transformers [J]. Science China Information Sciences, 2022, 65 (6): 160104.
[16]
TIANY, CHUX, WANGH. CCTrans: Simplifying and Improving Crowd Counting with Transformer [J]. ArXiv e-Prints, 2021: arXiv:
[17]
LIY, ZHANGX, CHEND. CSRNet: Dilated Convolutional Neural Networks for Understanding the Highly Congested Scenes [C]// 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2018: 1091-1100.
[18]
LIUW, SALZMANNM, FUA P. Context-Aware Crowd Counting [C]// 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2019: 5094-5103.
[19]
IDREESH, TAYYABM, ATHREYK, et al. Composition Loss for Counting, Density Map Estimation and Localization in Dense Crowds [M]// Computer Vision-ECCV 2018. Cham: Springer International Publishing, 2018, 11206: 544-559.
[20]
SINDAGIV A, YASARLAR, PATELV M. JHU-CROWD++: Large-Scale Crowd Counting Dataset and a Benchmark Method [J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2022, 44 (5): 2594-2609.
[21]
LIUL, LUH, ZOUH, et al. Weighing Counts: Sequential Crowd Counting by Reinforcement Learning [M]// Computer Vision-ECCV 2020. Cham: Springer International Publishing, 2020: 164-181.
[22]
MAZ, WEIX, HONGX, et al. Learning to Count via Unbalanced Optimal Transport [J]. Proceedings of the AAAI Conference on Artificial Intelligence, 2021, 35 (3): 2319-2327.
[23]
WANJ, LIUZ, CHANA B. A Generalized Loss Function for Crowd Counting and Localization [C]// 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). New York: IEEE Press, 2021: 1974-1983.