Current research in home and service robotics is largely focused on the development of smart robots that can perform tasks in real-world environments. In this context. vision-based scene understanding is considered to be an essential out highly challenging task due to several factors, including the segmentation of the visual information into object·hypotlileses and their recognition with the capability of generalization to unknown objects, under challenging conditions such as sensor noise, variable lighting conditions, occlusions and clutter. This thesis describes the development of novel computer vision algorithms for object-level scene understanding with high accuracy and real-time performance. We developed novel algorithms for 3D object segmentation. object recognition , and grasp detection using RGB·D imagery using low cost sensors. Our algorithms were Jested on challenging datasets and live videos acquired from a kinect mounted on our in-house robot named "AIPAR" to show the effectiveness and applicability or our algorithms in real-world environments.
|Qualification||Doctor of Philosophy|
|Award date||17 Aug 2016|
|Publication status||Unpublished - 2016|