南方科技大学知识苑(SUSTech KC): 基于垂直俯拍的区域人数统计视觉系统研究

Title	基于垂直俯拍的区域人数统计视觉系统研究
Alternative Title	RESEARCH ON VISUAL SYSTEM OF REGIONALPEOPLE COUNTING BASED ON VERTICALAERIAL PHOTOGRAPHY
Author	程嘉琪
Name pinyin	CHENG Jiaqi
School number	12032568
Degree	硕士
Discipline	0856 材料与化工
Subject category of dissertation	0856 材料与化工
Supervisor	刘伟
Mentor unit	机械与能源工程系
Publication Years	2022-05-12
Submission date	2022-06-25
University	南方科技大学
Place of Publication	深圳
Abstract	区域人数统计是指在通行出入口，通过传感和计算当时统计进出人数的实时数据，它不包括人的身份识别与行为分析。传统方法例，如红外对射探测、超声波技术、基于人脸/人体检测的安防视觉检测等，容易受到环境光、遮挡、隐私暴露等问题，在实际使用中还有很多障碍。本文研究以垂直角度的俯拍方式，融合彩色图像、3D 结构光点云，TOF 与双目视觉深度图像，处理俯拍视野下的人头顶部及肩膀区域数据，应用模式识别与深度学习的方法，实现区域人数统计视觉领域的改进与应用。本文采用垂直俯拍的硬件进行图像采集与预处理，接着进行多模图像的融合应用、行人的目标检测算法和行人追踪轨迹及区域的热区统计，然后基于嵌入式 AI 平台的系统部署和移植等问题进行深入研究。本人对具体细节提出了精度和性能优化方法，设计了完整的解决方案，搭建了完整的区域人数统计视觉系统并完成实地测试。具体的本文将从以下五个方面展开本项工作的论述： 1．本文分别从传统模式识别与深度学习在区域人数统计应用下的优缺点、区别等多个方面进行介绍对比。对比分析了人脸/人体检测与基于垂直俯拍的深度学习技术的优劣势。 2. 本文对硬件采集系统得到的多模图像进行预处理与融合分析，针对应用场景检测的行人特征点进行选取判断标注。 3．本文分别验证了两种高效的目标检测算法。在基于预权重的基础上进行训练，使用冷冻训练与解冻训练的优化方法，提高了系统的鲁棒性与准确率。使用训练模型在实地搭建验证了其实用性与准确度。 4．本文对目标场景下的错检情况的进行针对性改进，经过分析与论证，实地测试证明了改进方案格的可行性和优越性，并验证了移植嵌入式化终端的有效性。 5．本文具体分析了基于深度学习的目标追踪的原理，并对一段时间内的场景人流进行热力图保存分析。对出现在区域内行人的移动方向进行统计和判断，保存区域内行人的移动轨迹图。 6．基于以上研究基础，搭建了俯拍多模态头肩图像实验平台。该平台实现了较高准确率下对区域人数统计与客流分析，并可应用于多种复杂场景。在多个实地场景的测试中我得到不同改进模型适用范围，并计算分析了误差的产生原因。
Keywords	深度学习视觉检测目标追踪目标识别
Language	Chinese
Training classes	独立培养
Enrollment Year	2020
Year of Degree Awarded	2022-05
References List	[1] LEE D, CHA G, YANG M H, et al. Individualness and determinantal point processes for pedestrian detection[C]//European Conference on Computer Vision. [S.l.]: Springer, 2016: 330-346. [2] 周永虎. 视频流中基于头肩特征的运动人体检测与跟踪 [D]. [出版地不详]: 西安电子科技大学, 2013. [3] WU Y, YU T, HUA G. A statistical field model for pedestrian detection[C]//2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR’05): volume 1.[S.l.]: IEEE, 2005: 1023-1030. [4] TUZEL O, PORIKLI F, MEER P. Pedestrian detection via classification on riemannian manifolds[J]. IEEE transactions on pattern analysis and machine intelligence, 2008, 30(10): 1713-1727. [5] ZHIPING W. Pedestrian tracking algorithm based on human body characteristics identificationand kalman filter[J]. Electronics Optics & Control, 2016, 23(11): 97-102. [6] 应俊. 基于计算机视觉的电梯轿厢内人数统计研究 [D]. [出版地不详]: 杭州电子科技大学, 2013. [7] 冯化纲. 基于立体视觉的客流监控预警系统设计与实现 [J]. 计算机与数字工程, 2021, 49(4): 731-735. [8] 鲍华. 复杂场景下基于局部分块和上下文信息的单视觉目标跟踪 [D]. [出版地不详]: 中国科学技术大学, 2017. [9] GERONIMO D, LOPEZ A M, SAPPA A D, et al. Survey of pedestrian detection for advanceddriver assistance systems[J]. IEEE transactions on pattern analysis and machine intelligence,2009, 32(7): 1239-1258. [10] QIAN Y, LIANG J, PEDRYCZ W, et al. Positive approximation: an accelerator for attributereduction in rough set theory[J]. Artificial intelligence, 2010, 174(9-10): 597-618. [11] CHEN Y, ZHOU X S, HUANG T S. One-class svm for learning in image retrieval[C]//Proceedings 2001 International Conference on Image Processing (Cat. No. 01CH37205): volume 1. [S.l.]: IEEE, 2001: 34-37. [12] WANG X, HAN T X, YAN S. An hog-lbp human detector with partial occlusion handling[C]//2009 IEEE 12th international conference on computer vision. [S.l.]: IEEE, 2009: 32-39. [13] 肖军, 朱世鹏, 黄杭, 等. 基于光流法的运动目标检测与跟踪算法 [J]. 东北大学学报 (自然科学版), 2016, 37(6): 770. [14] DU X, EL-KHAMY M, LEE J, et al. Fused dnn: A deep neural network fusion approach to fastand robust pedestrian detection[C]//2017 IEEE winter conference on applications of computervision (WACV). [S.l.]: IEEE, 2017: 953-961. [15] 汪冲, 席志红, 肖春丽. 基于背景差分的运动目标检测方法 [J]. 应用科技, 2009, 36(10):16-18. [16] 刘琳. 基于人体头肩特征的行人检测方法研究与应用 [D]. [出版地不详]: 南京理工大学,2015. [17] RETTKOWSKI J, BOUTROS A, GÖHRINGER D. Real-time pedestrian detection on a xilinxzynq using the hog algorithm[C]//2015 International Conference on ReConFigurable Computing and FPGAs (ReConFig). [S.l.]: IEEE, 2015: 1-8. [18] BENENSON R, MATHIAS M, TIMOFTE R, et al. Pedestrian detection at 100 frames persecond[C]//2012 IEEE Conference on Computer Vision and Pattern Recognition. [S.l.]: IEEE,2012: 2903-2910. [19] 李春艳, 王立, 卢欣, 等. 一种双目立体视觉相机标定方法 [J]. 空间控制技术与应用, 2010,36(3): 51-54. [20] 邹朋朋, 张滋黎, 王平, 等. 基于共线向量与平面单应性的双目相机标定方法 [J]. 光学学报, 2018, 37(11): 1115006. [21] TANNER R, STUDER M, ZANOLI A, et al. People detection and tracking with tof sensor[C]//2008 IEEE Fifth International Conference on Advanced Video and Signal Based Surveillance.[S.l.]: IEEE, 2008: 356-361. [22] FELZENSZWALB P F, GIRSHICK R B, MCALLESTER D, et al. Object detection with discriminatively trained part-based models[J]. IEEE transactions on pattern analysis and machineintelligence, 2009, 32(9): 1627-1645. [23] 苏显渝, 张启灿, 陈文静. 结构光三维成像技术 [J]. 中国激光, 2014(2): 1-10. [24] GIRSHICK R, DONAHUE J, DARRELL T, et al. Rich feature hierarchies for accurate object detection and semantic segmentation[C]//Proceedings of the IEEE conference on computervision and pattern recognition. [S.l.: s.n.], 2014: 580-587. [25] DONG L, PARAMESWARAN V, RAMESH V, et al. Fast crowd segmentation using shapeindexing[C]//2007 IEEE 11th International Conference on Computer Vision. [S.l.]: IEEE, 2007:1-8. [26] HE K, ZHANG X, REN S, et al. Spatial pyramid pooling in deep convolutional networks forvisual recognition[J]. IEEE transactions on pattern analysis and machine intelligence, 2015, 37(9): 1904-1916. [27] VIOLA P, JONES M J, SNOW D. Detecting pedestrians using patterns of motion and appearance[J]. International Journal of Computer Vision, 2005, 63(2): 153-161. [28] QI C R, SU H, MO K, et al. Pointnet: Deep learning on point sets for 3d classification and segmentation[C]//Proceedings of the IEEE conference on computer vision and pattern recognition.[S.l.: s.n.], 2017: 652-660. [29] DU X, ANG M H, RUS D. Car detection for autonomous vehicle: Lidar and vision fusionapproach through deep learning framework[C]//2017 IEEE/RSJ International Conference onIntelligent Robots and Systems (IROS). [S.l.]: IEEE, 2017: 749-754. [30] KU J, MOZIFIAN M, LEE J, et al. Joint 3d proposal generation and object detection from viewaggregation[C]//2018 IEEE/RSJ International Conference on Intelligent Robots and Systems(IROS). [S.l.]: IEEE, 2018: 1-8. [31] 刘昊, 魏志强, 李臻, 等. 一种多相机系统的研究与实现 [J]. 中国海洋大学学报 (自然科学版), 2015, 3. [32] 常昕, 陈晓冬, 张佳琛, 等. 基于激光雷达和相机信息融合的目标检测及跟踪 [J]. 光电工程, 2019, 46(7): 180420-1. [33] 黄元捷. 基于随机蕨丛的改进型 TLD 跟踪算法 [J]. 计算机光盘软件与应用, 2015, 18(2):127-128. [34] 梁锡宁, 杨刚, 余学才, 等. 一种动态模板匹配的卡尔曼滤波跟踪方法 [J]. 光電工程, 2010,37(10): 29-33. [35] 钟必能. 复杂动态场景中运动目标检测与跟踪算法研究 [D]. [出版地不详]: 哈尔滨: 哈尔滨工业大学, 2010. [36] NGO V, CASADEVALL A, CODINA M, et al. A pipeline hog feature extraction for real-timepedestrian detection on fpga[C]//2017 IEEE East-West Design & Test Symposium (EWDTS).[S.l.]: IEEE, 2017: 1-6. [37] CHERKASSKY V, MA Y. Practical selection of svm parameters and noise estimation for svmregression[J]. Neural networks, 2004, 17(1): 113-126. [38] LI J, WANG H, ZHANG L, et al. The research of random sample consensus matching algorithm in pca-sift stereo matching method[C]//2019 Chinese Control And Decision Conference(CCDC). [S.l.]: IEEE, 2019: 3338-3341. [39] REDMON J, FARHADI A. Yolov3: An incremental improvement[J]. arXiv preprintarXiv:1804.02767, 2018. [40] 【必备】目标检测中的评价指标有哪些？[EB/OL]. (2020-08-01) [2021-06-04]. https://cloud.tencent.com/developer/article/1624811. [41] XING Y, LI A, CUI Z, et al. Moving target tracking algorithm based on improved kernelizedcorrelation filter[J]. Infrared and Laser Engineering, 2016, 45(s1): S126004. [42] AIZENBERG N N, AIZENBERG I N. Cnn based on multi-valued neuron as a model of associative memory for grey scale images[C]//CNNA’92 Proceedings Second International Workshopon Cellular Neural Networks and Their Applications. [S.l.]: IEEE, 1992: 36-41. [43] ZHANG H, ZHANG J, WU Q, et al. Extended kernel correlation filter for abrupt motion tracking[J]. KSII Transactions on Internet and Information Systems (TIIS), 2017, 11(9): 4438-4460. [44] TAN S, LIU Y, LI Y. Improved kernel correlation filter tracking with gaussian scale space[C]//Infrared Technology and Applications, and Robot Sensing and Advanced Control: volume10157. [S.l.]: SPIE, 2016: 713-719. [45] XU F, WANG H, SONG Y, et al. A multi-scale kernel correlation filter tracker with featureintegration and robust model updater[C]//2017 29th Chinese Control And Decision Conference(CCDC). [S.l.]: IEEE, 2017: 1934-1939. [46] MARIN J, VÁZQUEZ D, GERÓNIMO D, et al. Learning appearance in virtual scenariosfor pedestrian detection[C]//2010 IEEE computer society conference on computer vision andpattern recognition. [S.l.]: IEEE, 2010: 137-144.
Academic Degree Assessment Sub committee	创新创业学院
Domestic book classification number	TP751.1
Data Source	人工提交
Document Type	Thesis
Identifier	http://kc.sustech.edu.cn/handle/2SGJ60CL/342787
Department	School of Innovation and Entrepreneurship
Recommended Citation GB/T 7714	程嘉琪. 基于垂直俯拍的区域人数统计视觉系统研究[D]. 深圳. 南方科技大学,2022.

Files in This Item:
File Name/Size	DocType	Version	Access	License
12032568-程嘉琪-创新创业学院.（14901KB）			Restricted Access	--	Fulltext Requests

If you have any objections to this item, please fill out the form below and the administrator will contact you as soon as possible.
Content:
Email：	*
Affiliation Sequential No:
Verification Code:	Refresh

Any comments and suggestions are welcomed.
Title:	*
Content:
Email：	*
Verification Code:	Refresh