| As 3D technology, including computer graphics, virtual reality and 3D printing have been rapidly developed in the past years,3D models are gaining an increasingly huge demand. Traditional 3D modeling platforms such as Maya and ZBrush, utilize " windows, icons, menus, pointers " (WIMP) interface paradigms for fine-grained con-trol to construct perfect models. However the modeling progress can be verbose and frustrated and thus is too hard for a novice user or even a well-trained artist. Therefore a more intuitive interface is urgently needed.Sketch, an intuitive communication and modeling tool for human beings has be-come the first selection for modeling community.So far various sketch-based modeling systems have been created and studied. In this paper we attempt to show how these systems work and give a comprehensive sur-vey. We use the notion of pipeline and put forward our own analysis in a methodology perspective. We review and categorize the systems in four aspects:the input, the knowl-edge they use, the modeling approach and the output. We initially need the input, and with certain knowledge assisted, we use different approaches to achieve user-defined output such as scenes and models. This pipeline is more suitable for recent SBM work while agreeing with traditional approaches. We also discuss about inherent challenges and open problems for researchers in the future.There are two well-known survey papers which summarize those work compre-hensively. However, compared to them, this paper is still different and novel in two aspects. For the contents, we include the data-driven methods for sketch-based mod-eling. As for the recent decades, series of excellent methods in machine learning have been applied to SBM, and the paper thus offers a detailed analysis for those up-to-date work. On the other hand, for the categorization, previous surveys have put great efforts on specific techniques of SBM. Here we realize that the interpretation of sketch is based on not only the techniques they used, but also the knowledge behind. So this paper ad-ditionally makes a discussion on the knowledge part, providing a brand new point of view for analysis. |