BEGIN:VCALENDAR VERSION:2.0 METHOD:PUBLISH PRODID:-//Telerik Inc.//Sitefinity CMS 14.4//EN BEGIN:VTIMEZONE TZID:Mountain Standard Time BEGIN:STANDARD DTSTART:20231102T020000 RRULE:FREQ=YEARLY;BYDAY=1SU;BYHOUR=2;BYMINUTE=0;BYMONTH=11 TZNAME:Mountain Standard Time TZOFFSETFROM:-0600 TZOFFSETTO:-0700 END:STANDARD BEGIN:DAYLIGHT DTSTART:20230301T020000 RRULE:FREQ=YEARLY;BYDAY=2SU;BYHOUR=2;BYMINUTE=0;BYMONTH=3 TZNAME:Mountain Daylight Time TZOFFSETFROM:-0700 TZOFFSETTO:-0600 END:DAYLIGHT END:VTIMEZONE BEGIN:VEVENT DESCRIPTION:Bo Lang\, PhD\nElectrical Engineering\, University of Colorado Denver\nAbstract\nAutomatic recognition of human actions is an important a nd challenging problem in surveillance and intelligence transportation are as. Dynamics of human body skeletons convey significant information for hu man action recognition\, which attracted much attention in computer vision . Skeleton-based action recognition is widely used in recent applications due to its robustness to illumination and scene changes.\nThe skeletal mod ality can be represented naturally by a time series of body keyjoints with the form of 2D or 3D coordinates. Human motion can then be identified by analyzing its motion patterns. Conventional pose estimation approaches for getting skeleton data which are done on visible color imaging data could cause several problems in environments such as occlusion or complex backgr ounds and low illumination. Whereas thermal camera is stable to human body detection regardless of the lighting condition. On the contrary\, thermal data always lose the fine visual details of human objects\, especially at long distance. \nIn this seminar\, our multispectral pose estimation algo rithm used to generate the skeleton body keyjoints data from multispectral images or videos will be presented firstly. Then\, to capture richer depe ndencies besides the fixed skeleton graphs\, we develop the Two Steam Feed back Attention Based GCN (2S-ATGCN) model for skeleton-based action recogn ition which stacks attention-based graph convolution and temporal convolut ion as a basic building block\, to learn both spatial and temporal feature s from multispectral data in different environment. \;Experiments were conducted to evaluate both pose estimation and human action recognition p arts on the typical datasets. Results have showed that the overall model a chieves large improvement for human action recognition under the dark envi ronment. The flexibility of 2S-ATGCN model also opens up many possible dir ections for future works.\n \;\nEmail Address: bo.lang@ucdenver.edu DTEND:20191022T181500Z DTSTAMP:20240329T001325Z DTSTART:20191022T170000Z LOCATION: SEQUENCE:0 SUMMARY:CEDC seminar series: Multispectral skeleton-based human action reco gnition UID:RFCALITEM638472464053946604 X-ALT-DESC;FMTTYPE=text/html:
Bo Lang\, PhD
\nElectrical Engineering
\, University of Colorado Denver
Abstract
\nAutomatic recognition of human actions is an important and challenging pr oblem in surveillance and intelligence transportation areas. Dynamics of h uman body skeletons convey significant information for human action recogn ition\, which attracted much attention in computer vision. Skeleton-based action recognition is widely used in recent applications due to its robust ness to illumination and scene changes.
\nThe skeletal modality can be represented naturally by a time series of body keyjoints with the form of 2D or 3D coordinates. Human motion can then be identified by analyzing its motion patterns. Conventional pose estimation approaches for getting s keleton data which are done on visible color imaging data could cause seve ral problems in environments such as occlusion or complex backgrounds and low illumination. Whereas thermal camera is stable to human body detection regardless of the lighting condition. On the contrary\, thermal data alwa ys lose the fine visual details of human objects\, especially at long dist ance.
\nIn this seminar\, our multispectral pose estimation algorit hm used to generate the skeleton body keyjoints data from multispectral im ages or videos will be presented firstly. Then\, to capture richer depende ncies besides the fixed skeleton graphs\, we develop the Two Steam Feedbac k Attention Based GCN (2S-ATGCN) model for skeleton-based action recogniti on which stacks attention-based graph convolution and temporal convolution as a basic building block\, to learn both spatial and temporal features f rom multispectral data in different environment. \;Experiments were co nducted to evaluate both pose estimation and human action recognition part s on the typical datasets. Results have showed that the overall model achi eves large improvement for human action recognition under the dark environ ment. The flexibility of 2S-ATGCN model also opens up many possible direct ions for future works.
\n \;
\nEmail Address: bo.lang@ucdenver.edu
END:VEVENT END:VCALENDAR