BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Computer Vision Automated Productivity Measurement - Eirini Konsta
 ntinou - PhD
DTSTART:20141107T150000Z
DTEND:20141107T153000Z
UID:TALK55480@talks.cam.ac.uk
CONTACT:Lorna Everett
DESCRIPTION:Over the past decades\, the construction industry lags further
  and further behind the manufacturing sector when productivity is consider
 ed. This is due to inconsistency of internal factors that take place on si
 te\, which are mainly related to management and workforce issues. Almost a
 ll of them are directly related to the way that productivity is measured. 
 The currently applied methods for measuring productivity are labour intens
 ive\, time - cost consuming and error prone. They are mainly reactive moni
 toring processes initiated after the detection of a negatively influencing
  factor. The data collection is manual\, based on either work sampling or 
 reviewing surveillance video/photo data. Although research studies have be
 en performed towards leveraging these limitations\, a gap still exists in 
 extracting productivity rates for every construction entity and for all th
 e tasks that take place at a jobsite simultaneously\, without requiring pr
 ior knowledge regarding the type of task or work zone. In order to overcom
 e the aforementioned limitations\, the focus of this research is to propos
 e a method that will proactively measure productivity for the entire range
  of operations\, time and cost effectively. In general\, the aim is to aut
 omatically identify cycle patterns in trajectory data taken from jobsite
 ’s surveillance system. The goal of the first year of this project was t
 o develop a method that is capable of detecting repetitive patterns at job
 sites’ complex environments. For this purpose\, semantic analysis and tr
 ajectory analysis were implemented. Tasks are divided into cycle events\, 
 which are consisting of three semantic components. Two “stops” parts d
 epicting the execution of an activity and their in between connection with
  a “move” part. The initial results show that the former can be detect
 ed with density based clustering whereas the latter with curve simplificat
 ion algorithms. Future work\, will concentrate on 1) extracting “clean
 ” trajectory data\, registered on a single base in order to cover all ty
 pes of activities and also detect possible causes of delays\, 2) detection
  of cycle paths\, representing summary tasks consisting of subtasks (cycle
  events)\, 3) computation of productivity rate (cycles per hour) and 4) on
 line prediction of low productivity rates\, given training data of operati
 ons fixed in time and space. Such an approach will contribute to the impro
 vement of activities’ performance\, since productivity will be measured 
 proactively\, covering the entire range of operations on an individual lev
 el. Therefore\, the surveillance engineers will be provided with enough de
 tailed information to be able to proceed with the appropriate corrective a
 ctions.
LOCATION:Cambridge University Engineering Department\, LR5
END:VEVENT
END:VCALENDAR