Associating Drop and Pick Events in the Bicycles Domain

Dima Damen, David Hogg

This research aims at recognising drop and pick events in cluttered environments, as well as connecting drops with picks of the same object. This helps in providing an automatic log of events occurring. It can also be utilized in security applications like theft detection. Due to visual ambiguities arising from occlusion and noise, we find the best constrained globally optimal explanation given all data up-to-date. The approach is being evaluated on a several-days datasets from CCTV cameras.

Research Motivation

Several local councils had been installing CCTV cameras on top of bicycles racks to deter bicycle thefts. In June 2007, York had announced such a scheme after they noticed an alarming increase in bicycle thefts (The press). Yet installing CCTV cameras that are manually monitored has not proven as efficient as required in London (Daily Mail). A solution is thus sought to minimize the monitoring time and highlight cases where thefts are suspected.

Demos and Videos

Dataset Sample (AVI 3.7MB)
Correctly Detected Theft: Drop (AVI 4.2MB), Pick (AVI 5.0MB)
Online event logging and linking visualization (AVI 0.7MB)

Publications

Damen, Dima and Hogg, David (2009). Recognizing Linked Events: Searching the Space of Feasible Explanations. Computer Vision and Pattern Recognition (CVPR) pdf

Damen, Dima and Hogg, David (2007). Associating People Dropping off and Picking up Objects. British Machine Vision Conference (BMVC 07).

Damen, Dima and Hogg, David (2007). Bicycle Theft Detection. International Crime Science Conference. (CS2 07)

Media Coverage

Feb 2008: Bike chains get high-tech backup Briefing Latest, Engineering and Technology
Mar 2008: Safer cycles insight - The business magazine of the University of Leeds
Feb 2008: Silicon essentials - Bicycle Theft Detection HERO
Dec 21st 2007: Student's camera technology will help combat bicycle thefts. Yorkshire Post
Dec 20th 2007: Putting the breaks on bicycle thieves. Press Release by the University of Leeds PR office

The Method

(a)(b)
(a) When activity is detected inside the racks area, changed pixels grouped into clusters and trajectories are stochastically linked
(b) The match between pixels is used to connect drop to pick events
(c) A constrained globally optimal solution is seeked for all observations available until now

Acknowledgement

National Express East Anglia kindly assisted in the research by providing a dataset from one of their CCTV cameras overlooking a cluttered bicycle rack in a train station.