Engineering HDR Seminar Series No 3, 2015

14 January 2015, 15:00 – 16:00

Room 5-4-22, Monash University Malaysia

Abnormal human activity detection has been an active research area recently due to its wide range of applications for both public places and private properties. Often abnormal human activity detection aims to detect an abnormal human activity that contextually can be defined as suspicious, irregular, uncommon, anomaly etc. Abnormal activity detection can detect criminal activity, falling down of elders etc as these activities are rare events and occur infrequently but often causes damaging consequences. Three research areas are identified in abnormal human activity detection which are abnormality detection in single camera view, multi-camera views and RGB-Depth camera. Most challenging problem in these research areas is to find an feature descriptor with a lower computational cost and this issue is selected as our research focus for this study. To address this issue, we proposed a feature based on spatio-temporal features of an image sequence in a single camera scenario. Extraction of spatio-temporal features are based on modified 3-dimensional Harris function. Existing approaches such as optical flow are replicated to compare the performance of the proposed algorithm. Some initial investigation has also been conducted on an another recent approach known as mixture of dynamic texture model, to compare the performance and the computational cost. Compared to other state of the art descriptors with the UCSD dataset, our proposed descriptor has shown competitive performance with a lower computational cost based on initial findings. Abnormal human activity that appears in the scene can also be captured using multiple cameras.

Contact details:
Wan Nurul Rukiah Wan Rasdi
wan.nurul@monash.edu
Tel: 03-55146224