VisionBlocks: A Social Computer Vision Framework

Abstract

Vision Blocks (http://visionblocks.org) is an on demand, in-browser, customizable computer vision application publishing platform for masses. It empowers end-users (consumers)to create novel solutions for themselves that they would not easily obtain off-the-shelf. By transferring design capability to the consumers, we enable creation and dissemination of custom products and algorithms. We adapt a visual programming paradigm to codify vision algorithms for general use. As a proof of-concept, we implement computer vision algorithms such as motion tracking, face detection, change detection and others. We demonstrate their applications on real-time video. Our studies show that end users (non programmers) only need 50% more time to build such systems when compared to the most experienced researchers. We made progress towards closing the gap between researchers and consumers by finding that users rate the intuitiveness of the approach in a level 6% less than researchers. We discuss different application scenarios where such study will be useful and argue its benefit for computer vision research community. We believe that enabling users with ability to create application will be first step towards creating social computer vision applications and platform.