《(7.3)--第七章课外阅读：穿戴式相机.pdf》由会员分享，可在线阅读，更多相关《(7.3)--第七章课外阅读：穿戴式相机.pdf（16页珍藏版）》请在taowenge.com淘文阁网|工程机械CAD图纸|机械工程制图|CAD装配图下载|SolidWorks_CaTia_CAD_UG_PROE_设计图分享下载上搜索。
1、022VOLUME 7/ISSUE 2/APRIL 20191 研究背景当前，城市规划领域中对微观尺度的居民需求与空间品质的关注度不断提升，城市设计与更新也更加强调个体空间的使用与体验。而在研究人体行为与城市设计的关系方面，国内外学界均已有一定积累。自20世纪80年代起，芦原义信1、克莱尔库珀马库斯2、扬盖尔3等人陆续奠定了行为与场所空间使用之间关系的研究基础，劳伦斯D弗兰克等4及安妮瓦尔内兹穆东等5则基于实证研究，探索https:/doi.org/10.15302/J-LAF-20190203 收稿时间 RECEIVED DATE/2019-01-31 中图分类号/TP2,TU984 文献标
2、识码/A摘要 随着第四次工业革命的到来，人们开始不断探索新技术和新设备在研究个体行为与城市设计关系上的潜力。穿戴式相机的出现为监测个体在空间中的行为、形成个体“生命日志”提供了更多可能。本文以个体佩戴者为单位，探究穿戴式相机在研究个体行为与建成环境关系中的应用。研究通过人工识别、调用计算机视觉分析应用程序编程接口、利用Matlab进行色彩识别三种方式，对佩戴者在一周时间内收集到的8 598张照片进行了图像识别，并对比了三种方法的优劣势。随后，基于准确性较高的人工识别结果，研究对个体行为特征、时间分配、路径转移、场所事件等方面展开分析。研究表明，穿戴式相机采集的图片大数据库蕴含丰富的个体行为与时
3、空信息，可以有效描述个体在空间中的行为特征，对研究个体行为与建成环境之间的关系具有重要意义。关键词 可穿戴设备；时空间行为；生命日志；数字自我；图片大数据ABSTRACT With the advent of the Fourth Industrial Revolution,people have begun to explore the potential for new technologies and new devices in studying the relationship between human behavior and urban design.The emergence
4、 of wearable cameras offers more possibilities for monitoring individual behavior in built environments as a kind of“lifelog.”This article explores the applications of wearable cameras in studying the relationship between individual behavior and built environments.Using manual image identification,i
5、mage recognition with Computer Vision Application Programming Interface(API),and color calculation in Matlab,this study analyzed 8,598 photos recording the volunteers behaviors and activities during a week.Based on high-accuracy manual image identification results,the research analyzed the volunteer
6、s behavior,time use,movement path,and experiencing scenes.The study showed that the big data base of images collected by the wearable cameras contained rich individual activities and spatiotemporal information that could be used to effectively describe the individual behavior in space and further co
7、ntribute to the study of the relationship between individual behaviors and built environments.KEY WORDS Wearable Device;Spatiotemporal Behavior;Lifelog;Quantified Self;Big Data of Pictures编辑 汪默英 佘依爽 翻译 张健 萨拉雅各布斯 EDITED BY WANG Moying SHE Yishuang TRANSLATED BY Angus ZHANG Sara JACOBS张昭希清华大学建筑学院研究助理；
8、同济大学建筑学硕士龙瀛*清华大学建筑学院特别研究员、博士生导师ZHANG ZhaoxiResearch Assistant at School of Architecture,Tsinghua University;Master in Architecture,Tongji UniversityLONG YingSpecial Researcher and Doctoral Supervisor at School of Architecture,Tsinghua University本文引用格式/Please cite this article as:Zhang,Z.X.,&Long,Y.(
9、2019).Application of Wearable Cameras in Studying Individual Behaviors in Built Environments.Landscape Architecture Frontiers,7(2),22-37.https:/doi.org/10.15302/J-LAF-20190203穿戴式相机在研究个体行为与建成环境关系中的应用APPLICATION OF WEARABLE CAMERAS IN STUDYING INDIVIDUAL BEHAVIORS IN BUILT ENVIRONMENTS*通讯作者地址：北京市清华大学建
10、筑学院邮编：100084邮箱：DA19040249-2-p8-37-c6.indd 2219-5-7 上午7:08景观设计学/论文 LANDSCAPE ARCHITECTURE FRONTIERS/PAPERS0231 BackgroundNowadays,with the refinement of user needs and spatial quality,more attentions are paid to improve user experience in urban design and renewal projects.Also,scholars have made thei
11、r efforts in studying the relationship between individual behaviors and urban design.Since the 1980s,Yoshinobu Kuwahara1,Claire Cooper Marcus2,Jan Gehl3,and other pioneers have laid the foundation for the study on the relationship between human behaviors and space usage;Lawrence D.Frank et al.4,as w
12、ell as Anne Vernez Moudon et al.5,conducted empirical research on the impacts of environmental factors within urban blocks on human behaviors;Chinese scholars including Xu Leiqing et al.6,Pan Haixiao et al.7,and Chen Yong et al.8 have all focused on the changes of peoples behavior within the environ
13、ment by recording their activities in and interactions with the environment through observation,questionnaires,and interviews.Despite these achievements,few studies have focused on the behavioral characteristics of individuals in urban space.Limited by human labor and time cost,it has been difficult
14、 to objectively and continuously track and collect data of an objected individual in a long period of time.With the development of technology,the new data environment,formed by big data and open data,has provided a strong support for urban spatial research 9,including Point of Interest(POIs)data,bus
15、 IC card data10,and streetscape data11.Yet,the acquisition of individual data is still difficult.Exploring effective methods to obtain individual data will help make up for shortcomings in the current study on the interactions between individual behavior and built environment.2 New Device Developmen
16、t2.1 Wearable DevicesWith the dominance of digitization and intelligence,new technologies and new devices are emerging like mushrooms after rain.Common wearable devices such as smart bracelets and watches can record users physical condition and activities via human-computer interactions.At present,i
17、n the research of individual monitoring,there are two types of wearable devices mostly used:one is used to monitor bio-signals.For example,Amir Muaremi et al.12 used wearable chest belt to monitor individuals sleep quality and mental stress,and Peter Aspinall et al.13 demonstrated how mobile electro
18、encephalogram(EEG)can monitor individuals emotional changes.The other is used to record behaviors.For example,Park Jonghoon et al.14 used uni-(LC)and tri-axial accelerometers(AM,ASP)to analyze 了城市街区环境要素对人群活动的影响；徐磊青6、潘海啸7、陈泳8等中国学者则聚焦于人在环境中的行为变化，通过观察、问卷、访谈等方式记录了人们出行和接触环境的情况，分析了城市环境对人们行为的影响。尽管目前此类研究已经取
19、得一定成果，但关注个体使用者在城市空间中行为特征的研究仍然较少受人力和时间成本限制，研究者难以客观、长期、连续地对某一个体进行行为追踪和数据收集。随着技术的发展，基于大数据和开放数据共同构成的新数据环境已为新时代城市空间研究提供了有力支持9如高德或百度等导航地图中的兴趣点数据（POIs）、公交卡数据10、街景数据11等，但个体数据的获取仍然存在困难。探索获取个体数据的有效方法将有助于弥补目前在研究个体行为与建成环境互动关系上的短板。2 新设备发展2.1 可穿戴设备随着数字化和智能化的普及，新技术和新设备不断涌现。智能手环、智能手表等常见可穿戴设备已经可以通过人机交互记录佩戴者的身体状态和使用情
20、况。目前应用于个体监测研究中的可穿戴设备大致分为两类：一种以状态监测为主，如阿米尔穆阿雷米等12利用可穿戴胸带监测个体睡眠和心理压力情况，彼得阿斯皮纳尔等13利用心电传感器监测个体情绪变化；另一种以行为记录为主，如朴重勋等14利用计步器分析佩戴者的机体活动，乔治娜布朗等15利用穿戴式相机SenseCam帮助健忘症患者记录日常活动。1945年，范内瓦布什曾提DA19040249-2-p8-37-c6.indd 2319-5-7 上午7:08024VOLUME 7/ISSUE 2/APRIL 2019出“生命日志”的概念16，即使用智能设备全方位、全时段地记录个体生命特征，形成大量个体信息数据库，
21、以数字化的形式记录个体自我活动，从而促进人们对人与环境互动关系的了解这正是可穿戴设备的应用领域。目前，可穿戴设备已经被应用在医疗健康、环境感知、数据分析等领域。在中国，尽管其发展还处于新兴阶段，但也已被应用于多个学科的研究之中。例如范长军和高飞17从计算机科学的角度，尝试利用可穿戴传感器获取人体速度、生理信号等信息，并对深度神经网络识别人体活动的过程进行了讨论；冯雪和张丹18从社会心理学的角度提出利用移动传感技术采集心理和行为数据；虽然目前可穿戴设备在建成环境相关学科中的应用较少，但仍有学者在进行积极探索，如陈筝和刘颂19基于可穿戴传感器的情绪体验测量技术，采集人的生理信号，记录人在建成环境中
22、的情绪变化；申悦和柴彦威20通过可穿戴GPS设备记录居民的移动轨迹，测度居民的空间活动范围。2.2 穿戴式相机20世纪70年代，史蒂夫曼恩21基于人机交互概念研发出穿戴式相机。其后，随着微电子技术的发展，穿戴式相机也不断迭代更新，出现了SenseCam、GoPro、Narrative Clip等微型可携带相机，可提供记录个体活动的大量图片数据。瓦伊娃卡尼卡特等22发现由SenseCam相机拍摄的照片可有效增强个体对细节、情绪和偏好的记忆。姬玛威尔逊等23邀请18名5281岁的老人连续佩戴SenseCam相机一周，发现可穿戴设备比人们预想中更便于使用使用者会日渐习惯佩戴，且几乎不会对其生活造成影
23、响。亚伦杜安等24利用视觉检测器来识别穿戴式相机获得的图片数据中的特征要素，如电脑、打印机、记事本等，从而对个体行为进行记录。诸如此类的研究已证实了穿戴式相机在记录个体活动与“生命日志”方面的优势。另外也已有学者对穿戴式相机在建成环境中的应用展开了探索。例如，蒂莫西钱伯斯等25邀请了168名1113岁的新西兰儿童佩戴穿戴式相机，并辅以GPS设备记录他们的日常活动，收集其活动点分布及活动情况；安柏L皮尔逊等26利用穿戴式相机调查儿童日常亲水程度，通过分析每张图片中水体所占的像素比例和其他图片数据，识别出23种滨水活动并统计出佩戴者在滨水空间中停留的时间。尽管如此，利用穿戴式相机和图片分析来研究个
24、体行为与城市空间之间的关系仍是一个新兴课题。individuals physical activities,and Georgina Brown et al.15 used wearable SenseCam to help amnesiacs record daily activities.In 1945,Vannevar Bush proposed the concept of“lifelog,”16 which means using intelligent devices to record individual life characteristics comprehensively
25、 and continuously,to form a large number of individual databases which record ones activities in a digital way,in order to promote a better understanding of the interaction between humans and the environment.Nowadays,wearable devices have been applied to achieve this prospect.Wearable devices have b
26、een used in fields such as medical health,environmental perception,and data analysis.Although it is still at the initial development stage in China,Chinese researchers have begun to more actively engage wearable devices in various disciplines.Fan Changjun and Gao Fei17,for example,discussed from the
27、 perspective of computer science about the process of applying wearable sensors to obtain individuals movement information and physiological signals and identifying human activities by using the deep neural network;Feng Xue and Zhang Dan18 employed mobile sensing technology to collect psychological
28、and behavior data from the perspective of social psychology.Despite the applications in built environments are rare,scholars such as Chen Zheng and Liu Song19 assessed real-time in-situ environmental affective experience by recording individuals physiological signals obtained with wearable bio-senso
29、rs,and Shen Yue and Chai Yanwei20 recorded the movement trajectory of community residents with wearable GPS devices to learn their daily reaching realms.2.2 Wearable CamerasIn the 1970s,Steve Mann21 developed a wearable camera based on human-computer interactions.Since then,with the development of m
30、icroelectronics,wearable cameras have been continuously upgraded,bringing forth SenseCam,GoPro,Narrative Clip,etc.that have become common devices to record numerous image data of individual activities.Vaiva Kalnikait et al.22 found that photos taken by a SenseCam camera could effectively enhance an
31、individuals memory of details,emotions,and preferences.Gemma Wilson et al.23 invited 18 volunteers aged from 52 to 81 to wear SenseCam cameras for a week,and found that they gradually got used to wearing the cameras without inconvenience to their daily life,which revealed that wearable devices are m
32、ore acceptable than expected.Aaron Duane et al.24 recorded individuals behavior by identifying the elements such as computers,printers,and notebooks from the images collected DA19040249-2-p8-37-c6.indd 2419-5-7 上午7:08景观设计学/论文 LANDSCAPE ARCHITECTURE FRONTIERS/PAPERS0253 实验设计为了探索穿戴式相机在个体行为与建成环境关系研究中的应
33、用，本研究试图利用穿戴式相机采集图片数据，并采用人工识别、调用计算机视觉分析应用程序编程接口（API）、利用Matlab进行色彩识别三种方式识别图片信息。随后，针对个体行为特征、时间分配、路径转移、场所事件等方面展开分析。具体研究框架如图1所示。此次实验的预实验于2018年9月3日进行，正式实验时间为2018年10月814日（星期一至星期日），研究范围为测试者工作及生活的北京市海淀区清华园及周边地区。实验采用Narrative Clip二代穿戴式相机，每隔30秒拍摄一张照片，记录个体从清晨出门前至晚间到家后的所有经历（表1，2）。实验要求测试者将穿戴式相机佩戴在胸前，保证拍摄不受衣服、头发等物
34、体遮挡，且镜头方向与个体前进方向一致（图2）；另要求佩戴者每天卸下设备后及时导出照片并充电。实验结束后，测试者可删除涉及隐私的图片，随后由研究团队对剩余有效图片进行分析。由表2可知，在正常情况下，实验每日可获得1 2001 500张图片，一周累计获得8 00010 000张图片。with visual concept detectors.All these research has evidenced the advantage of wearable cameras in recording individual activities and generating“lifelogs.”Sch
35、olars have also explored the applications of wearable cameras in the built environment.For example,with the aid of GPS device,Timothy Chambers et al.25 asked 168 children in New Zealand aged from 11 to 13 to carry wearable cameras to record their daily activity patterns.Amber L.Pearson et al.26 used
36、 wearable cameras to investigate childrens daily access to water areas.By analyzing the image data such as the pixel ratio of water in the pictures,23 types of waterfront activities and corresponding time spent were identified.Nevertheless,applying wearable cameras and the image analyses to study th
37、e relationship between individual behavior and urban space is still an emerging research interest.3 Experiment DesignAs an exploration that studies the relationship between individual behaviors and built environments with wearable cameras,this research attempted to capture image data and identify im
38、age information through three methods manual image identification,image recognition with Computer Vision Application Programming Interface(API),and color calculation in Matlab.Researchers then analyzed the volunteers behavior,time use,movement path,and experiencing scenes.The research framework is s
39、hown in Figure 1.A pre-experiment was carried out on September 3,2018.The formal experiment was performed from October 8 to 14,2018(Monday to Sunday)in Tsinghua Campus and the surrounding areas in Haidian District,Beijing where the volunteer lives and works.The experiment used a Narrative Clip 2 wea
40、rable camera that takes a photo every 30 seconds,recording all kinds of the volunteers activities,from getting ready to work in the morning until arriving back home at night(Table 1,2).The volunteer was asked to wear the camera in front of her chest and ensure that the shooting was not blocked by cl
41、othes or hair and that the camera faced the same direction as the individual moves towards(Fig.2).The volunteer also needed to export the photos and charge the camera every day.Photos that involve privacy could be deleted by the volunteer before submitting the rest to the research team for analysis.
42、Averagely,1,200 to 1,500 images can be obtained per day,which makes a total of 8,000 to 10,000 images per week(Table 2).第一步 Step 1个体图片收集 Individual data collection个体佩戴实验 Wearing experiment新设备 New device新技术 New technology研究思路 Research method第三步 Step 3行为空间 Behavior and space第二步 Step 2信息识别 Information
43、identification人工识别 Manual image identification调用计算机视觉分析API Image recognition with Computer Vision API利用Matlab进行色彩识别 Color calculation in Matlab行为特征 Behavior生活方式与时间分配 Lifestyle and time use空间移动 Space场所驻留与路径跟随 Places and tracks行为信息 Behavior地点信息 Place时间信息 Time频率信息 Frequency叙事性 Narrative descriptions空间性
44、 Spatiality连续性 Continuity规律性 Regularity建成环境 Built environment个体时空行为 Individual spatiotemporal behavior第四步 Step 4未来讨论 Future discussion从感知到量化 From perception to quantification从群体到个体 From group to individual1 张昭希，龙灜1.研究框架1.Research frameworkDA19040249-2-p8-37-c6.indd 2519-5-7 上午7:08026VOLUME 7/ISSUE 2
45、/APRIL 2019表1：佩戴者个体信息Table 1:The volunteers information女Female26中国北京Beijing,China科研工作者Scientific researcher158 cm53 kg性别Gender年龄Age现居地Location职业Occupation身高Height体重Weight身体质量指数Body Mass Index(BMI)身体情况Body condition病史Illness history22.4健康Healthy无N/A表2：图片记录情况Table 2:Collected data预实验Pre-experiment正式实验
46、Formal-experiment2018-09-032018-10-082018-10-092018-10-102018-10-112018-10-122018-10-132018-10-149:30 22:308:00 22:307:30 23:308:00 24:007:30 21:307:30 21:3010:30 21:3010:00 24:001,4551,2831,4821,4231,3061,274531（设备电量不足）(Low battery)1,2601,4461,2721,4541,4091,2871,2545311,253晴Sunny多云Cloudy晴Sunny多云Cl
47、oudy晴Sunny多云Cloudy晴Sunny多云Cloudy日期Date佩戴时间Time period wearing the camera照片总数Total number of photos有效照片数Number of valid photos天气Weather注：1.由于相机会在没有任何指示灯提醒的情况下因电量不足而自动停止拍摄，因此某些时段拍摄照片会相对较少。2.随着使用次数的增加，设备的记录能力和续航能力均有所下降，对实验结果造成了一定影响。Notes:1.Fewer photos were taken when the camera was in a low-battery an
48、d turned off automatically without indication.2.The photography quality and battery life of the camera declined as the usage time increased,which somehow had an impact on the results.4 图片要素识别方法4.1 人工识别运用人工识别的方式识别图片信息主要是指通过对图片特征进行归纳和判断，确定每张图片发生的地点、时间和事件。人工识别的优点在于对图片的解读更具体，更能把握重点信息，也能够对佩戴者行为进行常识性推测，缺点
49、是耗时较长。表3以2018年10月9日收集到的图片信息为例，呈现了以人工识别方式对佩戴者全天行为的记录。但因佩戴设备电量不足，因而未记录到其晚间通勤以及夜间活动的情况。全天共拍摄了1 482张照片，其中有效图片1 454张。4 Image Identification/Recognition Methods4.1 Manual Image IdentificationManual image identification is to distinguish the location,time,and scene of each picture according to the image el
50、ements.It sees an advantage in accurately interpreting and grasping key image information to speculate the volunteers behavior and experiencing scene.But it usually takes a very long time.Table 3 is a record of behavior obtained through manual image identification,taking the photos collected on Octo