SUMMARY

This thesis proposal has been motivated by the interests of deploying adaptive and flexible mobile sensor networks to inspect large civil structures. In a mobile sensor network, each node is a sensor carrying robot that is able to move around some structure by utilizing appropriate adhesion mechanisms. This research seeks to develop vision-based methods for guiding mobile robot navigating on 3D structure and obtain geometric measurements.
To solve these two problems, a new scheme that utilizes the robot onboard camera and a pair of externally located stereo cameras is proposed. The method takes advantage of both local and global visual information gathered from multiple views to overcome the stringent constraints imposed on feasible paths of the robot imposed by the structure. The general vision measurement models for single and multiple views are provided. In addition, an iterative linear method of vision sensor calibration is proposed. A robot localization method using stereovision is developed. The camera and measurement models will be employed to investigate the effects of combining local and global visual information on the performance of mobile robot navigation as well as accuracy of structure reconstruction. It is expected that the proposed multiple view scheme will find many different applications in general contactless vision based measurement.