Open Collections

3D imaging for outdoor workspace monitoring

University of British Columbia

This research investigates the use of laser scanners to provide fast and accurate workspace information on excavators. The thesis first evaluates the use of laser scanners to track a mining shovel's dipper position, in order to extract its arm geometry variables in real-time. The low spatial resolution of laser scanners and the need for accurate initialization challenge the reliability and accuracy of laser-based object tracking methods. The first contribution of this work is in introducing a dipper tracking algorithm that provides an accurate, reliable and real-time performance. To accomplish this task, the proposed algorithm uses the shovel dipper's kinematic constraints and pose history in a Particle Filter to perform an efficient global search in the workspace. The result is then used to initialize an Iterative Closest Point algorithm that uses a geometrical model of the dipper to perform a local search and increase the accuracy of the positioning estimate. In the experiments performed on a mining shovel, the algorithm reliably tracked the dipper in real-time, and obtained a mean dipper positioning error of 6.7cm – 0.3% of the dipper range. In contrast to arm geometry, most workspace information cannot be extracted using a single laser scanner since it cannot scan the entire scene at a real-time frame-rate. In addition to having a real-time frame-rate, 3D images have to be accurate and reliable to be useful for workspace variable extraction. Hence, this work investigates whether and how such images can be provided using a laser scanner and an intensity camera. The slow and sparse results of depth sensors (e.g., laser scanner) make them ineffective when used alone. This work proposes a method for fusing depth sensors with an intensity camera to capture accurate and dense 3D images. The proposed fusion uses a super-resolution algorithm based on bilateral filtering to upsample small depth sensor images, with the help of high-resolution images from an intensity camera. The proposed method obtains more accurate results compared to the literature – an error reduction of up to 6.8x based on the Middlebury benchmark – with sharper and more realistic edge definition.

Text

2011-09-28

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/37670

Electrical and Computer Engineering

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/