Machine vision techniques are applied in a wide variety of industries around the world for quality inspection of the products. One area that is getting a signi cant amount of attention is pipelines. Pipe manufacturers strive to do fast and reliable quality control. Furthermore, pipeline users are justly concerned about the sustainability of their aging infrastructure. Timely pro-active inspection is, therefore, paramount in balancing maintenance cost over time.

Omnidirectional imaging systems are gaining world-wide attention due to their cost e ectiveness, fast calibration, and size factor. However, due to their high distortion, they are rarely used for pipe inspection applications. This research attempts to provide a comprehensive and cost-e ective solution for the characterization of the inner surface of a pipe using omnidirectional sensors. This solution includes detection, classi cation, and position-referencing of the visual surface defects, and nally reconstructing of the interior surface of the pipe in 3D.

First, a comprehensive study on the optimal spatial resolution of omnidirectional sensors for the pipe inspection applications is presented. This provides a guideline in selecting the omnidirectional sensor for achieving the highest resolution in the system. A mobile platform is also designed for inspecting the interior surface of the pipe using the selected omnidirectional sensor. To estimate the position of the sensor inside the pipe, a novel algorithm based on nonlinear optimization was designed. Furthermore, a new algorithm is presented for detecting and classifying the visible defects. In addition, a novel texture representation algorithm is presented to overcome the challenges of the visual odometry for the seamless textures ( e.g. PVC pipes). Finally, a complete platform for reconstructing the interior surface of a pipe in 3D is presented. The proposed platform for pipe characterization has been tested in a lab con guration, and satisfactory results have been achieved.