Defects such as surface corrosion in gas pipelines can lead to catastrophic failure and are a threat to life, property, and the viability of the line. Current techniques use magnetic flux sensors or ultrasonics. While effective, in practice the data produced is difficult to interpret and is easily confused with non-defect pipe features leading to false positives. Additionally, finding the location of defects requires expensive INS systems. We will investigate the following vision-based algorithms for in-pipe vehicles to augment existing sensor technologies and thereby overcome these weaknesses: a) 1D Visual Simultaneous Localization and Mapping (vSLAM) to estimate the vehicle pose, thereby reducing the need for expensive INS solutions. vSLAM will provide the pipe structure with registered appearance data. b) Location-registered evolution of pipe appearance over time by combining data from multiple runs. Most corrosion can therefore be detected by changes in the pipe appearance over time. c) Vision-based weld and defect detectors using data driven machine learning classifier techniques. d) Visualization methods using registered imagery to provide greater situational awareness to a human operator. We will evaluate our algorithms using a range of real pipes with different types of defects. Because of false positives, current methods need visual confirmation by a field team. Our approach will allow remote visual inspection without a field team.