Imaging and 3D modeling using a laser range sensor: application in the wood industry
Michel Robert
Patrick Hébert (Supervisor)
Problem: In the wood industry, numerous efforts have been invested in optimizing the profitability of sawmills. The general objective involves maximizing the value of the products for each tree which is processed. At the present time sawmills have sophisticated equipment to ensure an optimal transformation of the trees. The optimization systems use laser triangulation captors for the rapid acquisition of the 3D shape of logs in continuous motion. These sensors enable the control of the positioning of the logs and the opening of the saws in real time. The results of the analyses allow us to determine that the geometric model obtained can be improved.
Motivation: Studies which have been carried out using a sawmill wood transformation process simulator can potentially improve the geometric model produced by a laser range sensor. The results of our simulations confirm these conclusions and show that an error in the geometric model at the primary breakdown step has a considerable impact on the yield of the sawmill.
Approach: The method used consists in a re-engineering of the existing system in order to improve the quality of the 3D modeling chain of the sensor. The configuration of the sensor is composed of six synchronized cameras and four co-planar laser projectors. From this system, an analysis of the sources of error is carried out for the laser trace detection steps in the images, the calibration of the sensor and the reconstruction of the 3D model. The processing algorithms are then revised according to the importance of the errors to create a more precise model and eventually proposals are made for the calibration and/or acquisition of images. The results obtained are then validated with the help of reference geometric models. Finally, an evaluation of the impact and the improvement of the 3D model on the yield of the sawmill plant is carried out.
Challenges: The existing system produces a geometric model whose precision is insufficient and does not allow the yield objectives of the sawmill to be obtained. Furthermore, this system uses an explicit calibration which requires extrinsic parameters of the cameras and laser planes. Thus, the challenge of this project involves the revising of the calibration model used and the proposal of a more precise implicit calibration method. A detailed study of the system must permit the sources of measurement errors to be identified and allow the necessary corrections to be made in order to optimize the yield potential. Finally, the challenge also consists in developing a procedure for the evaluation of the quality of the sensor.
Applications: Modern sawmills consist of automated machines and optimization systems for the performance of the wood transformation operations in real time. Since triangulation sensors are widely used in existing optimization systems, an improvement of the 3D model would have a considerable impact on the yield of the sawmills for Canada’s wood industry.
Calendar: January 2001 - December 2002
Web reference: /~mrobert
Last modification: Sep 26 2007 10:43AM by mrobert


©2002-. Computer Vision and Systems Laboratory. All rights reserved