Research on Crowd Abnormal Behavior Detection Based on Improved SSD
-
摘要: 针对人群异常行为检测任务中存在的算法复杂度较高,重叠遮挡等带来的检测精度低等问题,本文提出一种基于改进SSD(Single Shot Multi-box Detector)的人群异常行为检测算法。首先采用轻量级网络MobileNet v2代替原始特征提取网络VGG-16,并通过可变形卷积模块构建卷积层来增强感受野,然后通过将位置信息整合到通道注意力中来进行特征增强,能够捕获空间位置之间的远程依赖关系,从而可以较好处理重叠遮挡问题。实验结果表明,本文提出的算法对人群异常行为具有较好的检测效果。Abstract: Aiming at the problems of high algorithmic complexity and low detection accuracy caused by overlapping occlusions in abnormal crowd behavior detection, this paper proposes an algorithm for crowd abnormal behavior detection based on an improved single-shot multi-box detector(SSD). First, the lightweight network MobileNet v2 was used to replace the original feature extraction network VGG-16, and a convolutional layer was constructed by a deformable convolution module to enhance the receptive field. Feature enhancement was performed by integrating the position information into the channel attention, which can capture long-range dependencies between spatial locations, allowing for better handling of overlapping occlusions. The experimental results show that the proposed algorithm has a good detection effect on abnormal crowd behavior.
-
Key words:
- deep learning /
- abnormal behavior detection /
- SSD network /
- deformable convolution /
- attention mechanism
-
表 1 基于改进SSD的消融实验
Table 1. Ablation experiment based on improved SSD
Methods Ped1 Ped2 Test speed /fps AUC/% Test speed /fps AUC/% VGG-16 21.96 61.98 20.36 63.81 MobileNet v2 27.92 65.26 27.02 70.25 MobileNet v2+Deformable Conv 26.68 69.81 25.73 77.98 MobileNet v2+Deformable Conv+CA 26.59 74.50 25.41 88.93 -
[1] HU Y. Design and implementation of abnormal behavior detection based on deep intelligent analysis algorithms in massive video surveillance[J]. Journal of Grid Computing, 2020, 18(2): 227-237. doi: 10.1007/s10723-020-09506-2 [2] 张欣, 齐华. 基于YOLOv4的人体异常行为检测算法研究[J]. 计算机与数字工程, 2021, 49(4): 791-796. doi: 10.3969/j.issn.1672-9722.2021.04.034ZHANG X, QI H. Research on human abnormal behavior detection algorithm based on YOLOv4[J]. Computer and Digital Engineering, 2021, 49(4): 791-796. doi: 10.3969/j.issn.1672-9722.2021.04.034 [3] 胡学敏, 陈钦, 杨丽. 基于深度时空卷积神经网络的人群异常行为检测和定位[J]. 计算机应用研究, 2020, 37(3): 891-895. https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ202003057.htmHU X M, CHEN Q, YANG L. Detection and localization of abnormal crowd behavior based on deep spatiotemporal convolutional neural network[J]. Application Research of Computers, 2020, 37(3): 891-895. https://www.cnki.com.cn/Article/CJFDTOTAL-JSYJ202003057.htm [4] Almazroey A A, Jarraya S K. Abnormal events and behavior detection in crowd scenes based on deep learning and neighborhood component analysis feature selection[C]//Proceedings of the International Conference on Artificial Intelligence and Computer Vision (AICV2020), 2020: 258-267. [5] MU Y L, ZHANG B. Abnormal event detection and localization in visual surveillance[C] //Communications, Signal Processing, and Systems, 2020: 1217-1225. [6] LIU W, Anguelov D, Erhan D, et al. SSD: Single shot multibox detector[C]//Proceedings of European Conference on Computer Vision, 2016: 21-37. [7] HOU Q B, ZHOU D Q, FENG J S. Coordinate attention for efficient mobile network design[C] //2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) 2021: 13708-13717. [8] Ali K, MOHAMMAD S M. Improved anomaly detection in surveillance videos based on a deep learning method[C]// 8th Conference of AI & Robotics and 10th RoboCup Iranopen International Symposium, 2018: 73-81. [9] Sandler M, Howard A, Zhu M, et al. MobileNetV2: Inverted Residuals and Linear Bottlenecks[C]//2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition, 2018: 4510-4520. [10] DAI J F, QI H Z, XIONG Y W, et al. Deformable Convolutional Networks[C]//2017 IEEE International Conference on Computer Vision (ICCV), 2017: 764-773. [11] Mahadevan V, LI W, Bhalodia V, et al. Anomaly detection in crowded scenes[C]//Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR) 2010: 1975-1981. [12] WU W H, CHEN M Y, LI J H, et al. Visual information based social force model for crowd evacuation[J]. Tsinghua Science and Technology, 2022, 27(3): 619-629. doi: 10.26599/TST.2021.9010023 [13] PANG G S, YAN C, SHEN C H, et al. Self-trained deep ordinal regression for end-to-end video anomaly detection[C]//2020 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 2020: 12170-12179.