To address the loop closure misjudgment problem due to visual interference, this paper proposes a loop closure detection algorithm that uses semantic information to verify candidates. The algorithm retrieves loop closure candidates using a visual bag-of-words model, and applies a verification method to eliminate the mismatches. The proposed verification method first extracts semantic nodes on the basis of semantic information in the scene, and then calculates the node descriptors containing neighboring information, so that we can accurately match the semantic nodes across images, reducing the sensitivity of the algorithm to dynamic objects. Subsequently, the algorithm constructs a relative position network based on matched semantic nodes from two views, and verifies loop closure candidates based on the similarity of network, which improves the robustness of the algorithm against perceptual aliasing. Experimental results show that the semantic position verification method significantly improves the detection accuracy of the visual bag-of-words model. Compared with other classic algorithms, the overall loop closure detection algorithm has achieved leading performances in both detection accuracy and computational efficiency.