To address the issues of limited scene configuration and sparse traditional visual feature points in a ventilation duct environment, a method for modeling and localization of pipes based on point and line features is proposed by using two monocular cameras and two-line scan lidars as the primary sensing devices. Firstly, the LaneNet network and an improved random sample consensus algorithm are utilized to extract four wall-edge line features. Then, geometric and spatial constraints are employed to filter the detection results of the line segment detector algorithm, obtaining two vertical line features at the pipe joints. Next, the pipe width and robot yaw angle are calculated by using the line scan lidars. The depth information of the line features at the pipe joints is recovered and solved to obtain the height of the pipe. By combining the camera projection equations, the world coordinates of the line segment endpoints are obtained, and the pipe height is estimated. Finally, a pipe map coordinate system is established, and the two-dimensional robot position and pipe length are estimated. The experimental results show that the relative positioning error is within 9. 8 cm, and the relative modeling error is within 2. 9 cm, which could meet the requirements for pipe modeling and localization during robot inspection operations in the ducts.