人群密度方面的综述.pdf

发布时间：2022-05-31 发布人：admin 分类：说明书资料大小：4.51M 资料格式：pdf 举报版权申诉

qq_31452127-10131042-4744302543399551965.pdf-第1页.png

第1页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第2页.png

第2页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第3页.png

第3页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第4页.png

第4页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第5页.png

第5页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第6页.png

第6页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第7页.png

第7页 / 共67页

qq_31452127-10131042-4744302543399551965.pdf-第8页.png

第8页 / 共67页

文本预览

Robert Collins Penn State Crowd Scene Analysis • Using computer vision tools to look at people in public places • Real-time monitoring – situation awareness – notifications/alarms • After-action review – traffic analysis VLPR 2012

Robert Collins Penn State Crowd Scene Analysis Things we might want to know: • How many people are there? • How to track specific individuals? • How to determine who is with whom? Challenges: Crowd scenes tend to have low resolution. You rarely see individuals in isolation. Indeed, there are frequent partial occlusions. VLPR 2012

Robert Collins Penn State Crowd Counting FAQ: How many people participated in ... • Tahrir Square Protests • Obama’s inaguration • Occupy Wall Street • Kumbh Mela VLPR 2012

Robert Collins Penn State Jacob’s Method • Herbert Jacobs, Berkeley, 1960s • count = area * density – 10 sqft/person – loose crowd (arm’s length from each other) – 4.5 sqft/person – more dense – 2.5 sqft/person – very dense (shoulder-to-shoulder) • Problem: Pedestrians do not uniformly distribute over a space, but clump together into groups or clusters. • Refinement: break area into a grid of ground patches and estimate a different density in each small patch. Accumulate these counts over whole area. VLPR 2012

Robert Collins Penn State Example of Jacob’s Method VLPR 2012 source http://www.popularmechanics.com/science/the-curious-science-of-counting-a-crowd

Robert Collins Penn State Computer Vision Could do Better! Cavaet: nobody really wants accurate counts e.g. organizers of the “Million Man March” in Washington DC threatened to sue the National Park Service for estimating that only 400K people attended. VLPR 2012

Robert Collins Penn State Vision-based Counting • detection and tracking (light density) • clustering feature trajectories that move coherently (moderate density) • treat crowd as a dynamic texture and compute regression estimates based on measured properties (heavy density) VLPR 2012

Robert Collins Penn State Detecting and Counting Individuals Ge and Collins, "Marked Point Processes for Crowd Counting," IEEE Computer Vision and Pattern Recognition (CVPR'09), Miami, FL, June 2009, pp.2913-2920. Good for low-resolution / wide-angle views. Relies on foreground/background segmentation. Not appropriate for very high crowd density or stationary people. VLPR 2012

分享到：

赞收藏

资料库

人群密度方面的综述.pdf

相关推荐

人工智能

热门标签

最新资料