Sketches are a ubiquitous form of communication in engineering design due to their simplicity and efficiency. However, because of a lack of suitable machine-interpretation techniques, they are virtually unusable with current computer-aided design and engineering tools. The informal nature of sketches and their inherent ambiguity present a number of challenges to the development of such techniques. Here we address one particular challenge, the task of reliably locating and recognizing the intended visual objects from a continuous stream of pen strokes. We present an integrated sketch parsing and recognition approach, based on a novel mark-group-recognize paradigm, which is tailored to the domain of mechanical systems. In the first step of processing, the sketch is examined to identify certain delimiting symbols called “markers.” The remaining pen strokes are then partitioned into distinct clusters, each representing a single symbol. Finally, a trainable symbol recognizer is used to find the best interpretation of each cluster. We have used these techniques to build a sketch-based tool for designing and analyzing vibratory mechanical systems. This tool enables designers to analyze and animate vibratory systems by simply sketching them on a tablet computer. User studies indicate that even first-time users find our tool to be effective.
Levent Burak Kara, Leslie Gennari, Thomas F Stahovich. (2008). A Sketch-Based Tool for Analyzing Vibratory Mechanical Systems. Journal of Mechanical Design, Volume 130, Issue 10.