Intelligent Statistic Method for Video Pedestrian Flow Considering Small Object Features
-
摘要:
人流量统计对智能安防等领域具有重要研究价值. 针对视频监控系统中人流量统计准确率较低的问题，提出一种顾及小目标特征的视频人流量智能统计方法，重点研究用于小目标检测的Faster R-CNN （Faster region-convolutional neural network）改进算法、运动目标关联匹配以及双向人流量智能统计等关键技术，根据人头目标的小尺度特点对Faster R-CNN网络结构进行适应性改进，利用图像浅层特征提高网络对于小目标的特征提取能力，通过基于轨迹预测的跟踪算法实现运动目标的实时追踪，同时设计双向人流量智能统计算法，以实现人流量高准确率统计；为证明方法的有效性，在密集程度不同的场景中进行了测试. 测试结果表明：改进的目标检测算法相较于原始算法，在Brainwash测试集和Pets2009基准数据集上的平均准确率分别提高了7.31%、10.71%，视频人流量智能统计方法在多种场景下的综合评价指标
F 值均能达到90.00%以上，相较于SSD-Sort算法和Yolov3-DeepSort算法，其F 值提高了1.14% ~ 3.04%.-
关键词:
- 监控视频 /
- 人流量统计 /
- Faster R-CNN改进算法 /
- 目标检测 /
- 目标跟踪
Abstract:Pedestrian flow statistics have an important value of research in the fields like intelligent security. In view of the low accuracy in pedestrian flow statistics of video surveillance systems, an intelligent statistic method of video pedestrian flow is proposed with small object features considered. Key technologies are focused in this work such as the improved Faster R-CNN (Faster region-convolutional neural network) algorithm for small object detection, moving object association and matching, and intelligent statistics of bidirectional pedestrian flow. More efforts include adapting the Faster R-CNN network structure according to the small-scale characteristics of the head object, improving the feature extraction ability of the network for small objects by using shallow features of images, and realizing real-time tracking of moving objects through the tracking algorithm based on trajectory prediction. Meanwhile, an intelligent statistic algorithm for bidirectional pedestrian flow is developed to achieve accurate statistics of pedestrian flow. To prove the effectiveness of the proposed method, the experiments were conducted in scenes with different levels of density. The results show that compared with the original algorithm, the improved object detection algorithm improves the mean average precision by 7.31% and 10.71% on the Brainwash test set and Pets2009 benchmark data set, respectively. For the intelligent statistic method of video pedestrian flow, the comprehensive evaluation index
F value in various scenes can reach above 90.00%, which is 1.14%−3.04% higher than the excellent methods in recent years. -
图 5 双向人流量智能统计算法流程
Figure 5. Flow chart of intelligent statistic algorithm of bidirectional pedestrian flow
图 8 不同场景下视频人流量统计结果
Figure 8. Results of video pedestrian flow statistics in different scenes
表 1 人流量统计实验视频信息
Table 1. Experimental video information of pedestrian flow statistics
视频序列 时长/s 帧率/（帧 · s−1） 人数 密集程度 视频一 222 25 65 稀疏 视频二 35 25 9 稀疏 视频三 46 30 63 密集 
下载: 导出CSV
表 2 不同目标检测算法结果对比
Table 2. Result comparison of different object detection algorithms
下载: 导出CSV
表 3 模型对比结果
Table 3. Result comparison of models
算法 平均准确率/% 时间/s 原始 63.53 0.166 对比 67.19 0.184 本文 74.24 0.168
下载: 导出CSV
表 4 多尺度特征融合方法精度
Table 4. Mean average precision of multi-scale feature fusion methods
Conv2_2 Conv3_3 Conv4_3 Conv5_3 平均准确率/% √ √ √ 64.92 √ √ √ 68.43 √ √ √ 69.46
下载: 导出CSV
表 5 视频人流量统计结果
Table 5. Results of video pedestrian flow statistics
% 视频序列 召回率 精确率 F 值 视频一 90.77 93.65 92.19 视频二 88.89 100.00 94.12 视频三 92.06 93.55 92.80
下载: 导出CSV
表 6 不同算法对比结果
Table 6. Comparison results of different algorithms for pedestrian flow statistics
% 算法 召回率 精确率 F 值 SSD-Sort 87.59 91.60 89.55 Yolov3-DeepSort 89.78 93.18 91.45 本文方法 91.24 93.98 92.59
下载: 导出CSV
-
[1] 鲁工圆,马驷,王坤,等. 城市轨道交通线路客流控制整数规划模型[J]. 西南交通大学学报,2017,52(2): 319-325. doi: 10.3969/j.issn.0258-2724.2017.02.015LU Gongyuan, MA Si, WANG Kun, et al. Integer programming model of passenger flow assignment for congested urban rail lines[J]. Journal of Southwest Jiaotong University, 2017, 52(2): 319-325. doi: 10.3969/j.issn.0258-2724.2017.02.015 [2] 张君军,石志广,李吉成. 人数统计与人群密度估计技术研究现状与趋势[J]. 计算机工程与科学,2018,40(2): 282-291. doi: 10.3969/j.issn.1007-130X.2018.02.013ZHANG Junjun, SHI Zhiguang, LI Jicheng. Current researches and future perspectives of crowd counting and crowd density estimation technology[J]. Computer Engineering & Science, 2018, 40(2): 282-291. doi: 10.3969/j.issn.1007-130X.2018.02.013 [3] XIE Y K, ZHU J, CAO Y G, et al. Efficient video fire detection exploiting motion-flicker-based dynamic features and deep static features[J]. IEEE Access, 2020, 8: 81904-81917. doi: 10.1109/ACCESS.2020.2991338 [4] 蔡泽彬. 基于视频分析的行人检测及统计方法研究[D]. 广州: 华南理工大学, 2015. [5] 李航,张涛,李菲. 一种基于智能视频分析的人流量统计算法[J]. 信息工程大学学报,2018,19(3): 373-378. doi: 10.3969/j.issn.1671-0673.2018.03.023LI Hang, ZHANG Tao, LI Fei. Pedestrian volume counting algorithm based on intelligent video analysis[J]. Journal of Information Engineering University, 2018, 19(3): 373-378. doi: 10.3969/j.issn.1671-0673.2018.03.023 [6] 徐超,高梦珠,查宇峰,等. 基于HOG和SVM的公交乘客人流量统计算法[J]. 仪器仪表学报,2015,36(2): 446-452.XU Chao, GAO Mengzhu, ZHA Yufeng, et al. Bus passenger flow calculation algorithm based on HOG and SVM[J]. Chinese Journal of Scientific Instrument, 2015, 36(2): 446-452. [7] 彭山珍,方志军,高永彬,等. 基于多尺度全卷积网络特征融合的人群计数[J]. 武汉大学学报(理学版),2018,64(3): 249-254.PENG Shanzhen, FANG Zhijun, GAO Yongbin, et al. Crowd counting based on feature fusion of multi-scale fully convolutional networks[J]. Journal of Wuhan University (Natural Science Edition), 2018, 64(3): 249-254. [8] CHAN A B, LIANG Z J, VASCONCELOS N. Privacy preserving crowd monitoring: Counting people without people models or tracking[C]//IEEE Conference on Computer Vision and Pattern Recognition. Anchorage: IEEE, 2008: 1-7. [9] ZHANG C, LI H, WANG X, et al. Cross-scene crowd counting via deep convolutional neural networks[C]// IEEE Conference on Computer Vision and Pattern Recognition. Boston: IEEE, 2015: 833-841. [10] 秦方. 基于计算机视觉的行人检测与人数统计算法研究[D]. 成都: 电子科技大学, 2018. [11] HE M, LUO H, HUI B, et al. Pedestrian flow tracking and statistics of monocular camera based on convolutional neural network and Kalman filter[J]. Applied Science-Basels, 2019, 9(8): 1-13. [12] 曹诚,卿粼波,韩龙玫,等. 城市量化研究中视频人流统计分析[J]. 计算机系统应用,2018,27(4): 88-93.CAO Cheng, QING Linbo, HAN Longmei, et al. Human traffic analysis based on video for urban quantitative research[J]. Computer Systems & Applications, 2018, 27(4): 88-93. [13] 张天琦. 基于深度学习的行人流量统计算法研究[D]. 哈尔滨: 哈尔滨工业大学, 2017. [14] REN S, HE K, GIRSHICK R, et al. Faster R-CNN: Towards real-time object detection with region proposal networks[J]. IEEE Transactions on Pattern Analysis and Machine Intelligence, 2017, 39(6): 1137-1149. doi: 10.1109/TPAMI.2016.2577031 [15] ZHANG L L, LIN L, LIANG X D, et al. Is faster R-CNN doing well for pedestrian detection?[M]// Computer Vision-ECCV 2016. Cham: Springer International Publishing, 2016: 443-457. [16] GATYS L A, ECKER A S, BETHGE M. Image style transfer using convolutional neural networks[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 2414-2423. [17] SIMONYAN K, ZISSERMAN A. Very deep convolutional networks for large-scale image recognition[EB/OL]. (2014-09-04)[2020-05-20]. https://arxiv.org/abs/1409.1556. [18] STEWART R, ANDRILUKA M, NG A Y. End-to-end people detection in crowded scenes[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 2325-2333. [19] BEWLEY A, ALEX Z, OZZ L, et al. Simple online and realtime tracking[C]//IEEE International Con- ference on Image Processing. Phoenix: IEEE, 2016: 3464-3468. [20] FERRYMAN J, SHAHROKNI A. PETS2009: Dataset and challenge[C]//Twelfth IEEE International Work- shop on Performance Evaluation of Tracking and Surveillance. Snowbird: IEEE, 2009: 1-6. [21] 吕俊奇,邱卫根,张立臣,等. 多层卷积特征融合的行人检测[J]. 计算机工程与设计,2018,39(11): 3481-3485.LYU Junqi, QIU Weigen, ZHANG Lichen, et al. Multi-scale convolutional feature fusion for pedestrian detection[J]. Computer Engineering and Design, 2018, 39(11): 3481-3485. [22] VORA A, CHILAKA V. FCHD: A fast and accurate head detector[DB/OL]. (2018-09-24)[2020-05-22].https://arxiv.org/abs/1809.08766v1. [23] 高玮军,师阳,杨杰,等. 一种改进的轻量人头检测方法[J]. 计算机工程与应用,2021,57(1): 207-212.GAO Weijun, SHI Yang, YANG Jie, et al. An improved lightweight head detection method[J]. Computer Engineering and Applications, 2021, 57(1): 207-212. [24] BELL S, ZITNICK C L, BALA K, et al. Inside-Outside Net: Detecting objects in context with skip pooling and recurrent neural networks[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 2874-2883. [25] KONG T, YAO A, CHEN Y, et al. HyperNet: Towards accurate region proposal generation and joint object detection[C]//IEEE Conference on Computer Vision and Pattern Recognition. Las Vegas: IEEE, 2016: 845-853. [26] 李继秀,李啸天,刘子仪. 基于SSD卷积神经网络的公交车下车人数统计[J]. 计算机系统应用,2019,28(3): 51-58.LI Jixiu, LI Xiaotian, LIU Ziyi. Statistics on number of people getting off bus based on SSD convolutional neural network[J]. Computer Systems & Applications, 2019, 28(3): 51-58. [27] 杨亦乐,高玮玮,马晓峰,等. 基于深度学习的行人数量统计方法[J]. 软件,2019,40(11): 119-122,151. doi: 10.3969/j.issn.1003-6970.2019.11.026YANG Yile, GAO Weiwei, MA Xiaofeng, et al. Pedestrian statistics based on deep learning[J]. Computer Engineering & Software, 2019, 40(11): 119-122,151. doi: 10.3969/j.issn.1003-6970.2019.11.026 -
下载:
点击查看大图
计量
- 文章访问数: 421
- HTML全文浏览量: 140
- PDF下载量: 49
- 被引次数: 0