Emerging Tech: VideoTrace

Creating great looking real-time ready models of complex objects is getting easier and easier.

Anton van den Hengel of the Australian Centre for Visual Technologies is leading the development of an amazing new process that allows you to easily create 3d models from video. The process - called 'VideoTrace' - debuted last year at SIGGRAPH and involves interactively tracing the desired object over several frames of video.

Check out a very cool demo video here or here.

Image processing algorithms have existed for years that attempt to automatically detect and generate 3d meshes from a set of images. Examples of successful implementations are found in medicine and geology, where data sets involve known components with little variation. The limiting factor in taking this idea to general use has primarily been the ambiguities (jitter, fade, artifacts, etc.) of an undefined visual source.

VideoTrace compensates for these shortcomings by providing an interface for the user to interactively trace the desired object in the video. These tracings give hints to the algorithms about what should be processed and (theoretically) remove the major ambiguities. A paper describing the technology can be found here.

A note on the ACVT VideoTrace page says that they are currently looking for support for commercial development. I sure hope they get it. Even though models of some small objects would still be easier to generate with a low-end 3d scanner, the applications for this technology on a large scale are huge. Think of a bridge between Google Streetwise and Google Earth - users could quickly trace the video of all the houses on their block and have it uploaded for use in Google Earth.... The possibilities are many!