3D point cloud semantic segmentation is an important frontier problem in the field of computer vision. It has a wide range of practical applications, especially in the fields of autonomous driving and robot perception. In recent years, the rapid development of deep learning technology has made cross-modal data fusion an effective means to improve the accuracy of point cloud semantic segmentation, especially the combination of point cloud data and visible light images. This fusion method can effectively cope with various challenges in complex environments. This paper takes cross-modal 3D point cloud semantic segmentation as the research background, discusses the fusion mode of visible light images and point cloud data, introduces and analyzes the existing fusion schemes in detail, and introduces the advantages and disadvantages of different fusion schemes and fusion efficiency. Then, this paper introduces the commonly used 3D point cloud semantic segmentation datasets and evaluation criteria. Since point cloud data can be extended to different data forms and is crucial to the fusion efficiency of modal fusion, this paper classifies the current mainstream algorithms according to different forms of point cloud data and discusses their efficiency. Finally, the future development direction of this field is prospected.