dc.contributor.author | S´anchez Matilla., Ricardo. | |
dc.date.accessioned | 2021-07-02T10:18:10Z | |
dc.date.available | 2021-07-02T10:18:10Z | |
dc.date.issued | 2021-03-30 | |
dc.identifier.citation | S´anchez Matilla., Ricardo. 2021. Object localisation, dimensions estimation and tracking. Queen Mary University of London. | en_US |
dc.identifier.uri | https://qmro.qmul.ac.uk/xmlui/handle/123456789/72858 | |
dc.description | PhD Theses. | en_US |
dc.description.abstract | Localising, estimating the physical properties of, and tracking objects from audio and video
signals are the base for a large variety of applications such as surveillance, search and rescue,
extraction of objects’ patterns and robotic applications. These tasks are challenging due to low
signal-to-noise ratio, multiple moving objects, occlusions and changes in objects’ appearance.
Moreover, these tasks become more challenging when real-time performance is required and
when the sensor is mounted in a moving platform such as a robot, which introduces further problems
due to potentially quick sensor motions and noisy observations. In this thesis, we consider
algorithms for single and multiple object tracking from static microphones and cameras, and
moving cameras without relying on additional sensors or making strong assumptions about the
objects or the scene; and localisation and estimation of the 3D physical properties of unseen objects.
We propose an online multi-object tracker that addresses noisy observations by exploiting
the confidence on object observations and also addresses the challenges of object and camera motion
by introducing a real-time object motion predictor that forecasts the future location of objects
with uncalibrated cameras. The proposed method enables real-time tracking by avoiding computationally
expensive labelling procedures such as clustering for data association. Moreover,
we propose a novel multi-view algorithm for jointly localising and estimating the 3D physical
properties of objects via semantic segmentation and projective geometry without the need to use
additional sensors or markers. We validate the proposed methods in three standard benchmarks,
two self-collected datasets, and two real robotic applications that involve an unmanned aerial vehicle
and a robotic arm. Experimental results show that the proposed methods improve existing
alternatives in terms of accuracy and speed. | en_US |
dc.language.iso | en | en_US |
dc.publisher | Queen Mary University of London. | en_US |
dc.title | Object localisation, dimensions estimation and tracking. | en_US |
dc.type | Thesis | en_US |
rioxxterms.funder | Default funder | en_US |
rioxxterms.identifier.project | Default project | en_US |