Sewer systems, as an important part of urban public infrastructure, require regular inspection of the pipelines to maintain normal function. Due to the special characteristics of the sewer environment, it is difficult to detect the situation inside the pipe with existing technologies. A robot localization system for sewer was proposed, which can locate the position of the robot and pipe defects. To address the challenges posed by sewers, which are often high or filled with water, a powered sonar robot was used and the UKF-RTS algorithm was introduced. This algorithm fuses data from two sensors of IMU(inertial measurement unit) and odometry to calculate the displacement and attitude of the robot. Experiments were designed to validate the system. The result shows that this system can accurately locate the position of the robot in the pipe, precisely calculate the position of defects in the pipe with an accuracy of up to 4.6%. It can achieve higher accuracy compared with single-sensor systems and other fusion algorithms, and combined with the 2D sonar point cloud data, it can generate a 3D point cloud model of the pipe, which has a high practical value.