From Data to Decision: Boosting for Time-Constrained Detection of Objects in Video
Gary Overett (ANU)
COMPUTER VISION AND ROBOTICS SERIESDATE: 2011-02-17
TIME: 16:00:00 - 17:00:00
LOCATION: RSISE Seminar Room, ground floor, building 115, cnr. North and Daley Roads, ANU
CONTACT: JavaScript must be enabled to display this email address.
ABSTRACT:
In this seminar I will attempt to discuss "my thesis in pictures and graphs". The thesis considered the detection of humans and road signs in video with significant ego-motion. The work of the thesis aimed to improve an object detection system in terms of both robustness and time-to-decision. The work covers numerous details and improvements to the popular boosted cascade of classifiers approach. Along the way worked to improve the quality of training data for a pedestrian detector by creating the NICTA Pedestrian Dataset which was at the time it was presented the largest pedestrian dataset available. We improve the weak learners used in boosting as well as creating new features with excellent discriminative power in tandem with low computational complexity. Given enough time I will also attempt to cover how some of this research fits into the work of the Automap project which has already used this research to detect road signs in over 150,000 kms of video.
BIO:
Gary Overett is currently a Research Engineer at NICTA CRL working on the Automap project. Gary first came to the RSISE in 2000 as a college student working on a simple robotics research project. In 2002-2004 while completing his software engineering degree at the ANU he took up part time work at RSISE/NICTA*. After a year in Kunming, China he joined NICTA as a PhD student.
Gary's most recent work is largely focussed on time-constrained sign detection in video. His interests are in machine learning (especially boosting) and fast and efficient features for object detection. In his increasingly sparse free time Gary enjoys windsurfing and hiking.
