Semantic concept detection in video is a challenge for video semantic content analysis. The performance of semantic concept detection methods depends on modeling and matching the video semantic content exactly. In this research, perception concept and semantic concept were defined to abstract and model video semantic content. Furthermore, the knowledge-assisted framework for semantic concept detection was proposed, in which the context knowledge was modeled using ontology, and the semantic concepts were detected by combining with low-level features and context information. Finally, the linear fusion strategy was used to fuse the matching results and detect the semantic concepts. The proposed method was demonstrated in a news video domain and shows promising results.