At present, the real-time localization technology in simultaneous localization and mapping (SLAM) algorithm has become stable, and the research field turns to the semi-dense SLAM based on point-line feature. Aiming at this research direction, a SLAM algorithm based on point-line feature is proposed. First, an oriented fast and rotated brief (ORB) algorithm based on three patches and local gray difference is used in the visual front end to extract and match feature points. Meanwhile, multiscale line segmentation detector (MLSD) algorithm is used to extract and match line segment features, so that the system adds constraint conditions of line features on the basis of point feature geometric transformation to calculate pose transformation. Then the position and orientation are optimized using point and line constraints through the local bundle adjustment (BA) method. Finally, loop closure detection is performed for repositioning. The proposed algorithm is tested on Euroc dataset and compared with similar algorithms. The experimental results show that the map is relatively dense, with clear outline and high accuracy. At the same time, the root mean square error (RMSE) in V1-02-mdeium, V2-02-mdeium, MH-02-easy, MH-03-medium, MH-04-difficult datasets is 0.045, 0.0561, 0.0539, 0.0491, 0.0623 respectively, which is relatively the lowest. The results show that the proposed algorithm has relatively good mapping effect and high accuracy among similar algorithms.