Robot Vision System Based on Log-Polar Image Plane Coordinates

SBIR二期项目的目标是设计和构建动态相机控制系统和处理硬件，以实现高性能实时视觉算法。高性能是基于对数极图像平面坐标的重构算法，这大大减少了像素计数和每像素的计算量。该系统的能力将包括基于双目立体和动态环境中的光流的深度感知。该系统可以作为视觉研究人员的测试平台进行复制。双目摄像机运动头将作为一个独立的模块，可以连接到任何处理架构。第一阶段展示了在对数极坐标下制定的完整的图像处理算法，证明了该系统的可行性和效率。计算需求的数量级降低表明机器人视觉的巨大优势，特别是移动机器人。该技术可以将机器人视觉系统的性价比提高几个数量级。这打开了传统的基于笛卡尔坐标的计算机视觉无法实现的商业利基。市场包括自动地板清洁工、运输车、安保机器人和服务机器人。

The objective of this SBIR Phase II project is to design and build a dynamic camera control system and processing hardware to exercise high performance real time vision algorithms. High performance is based on reformulating algorithms in log-polar image plane coordinates, which significantly reduces pixel count and computations per pixel. Capabilities of this system will include depth perception based on binocular stereo and optic flow in dynamic environments. This system can be duplicated as a testbed for vision researchers. The binocular camera motion head will be available as a self-contained module which can be attached to any processing architecture. Phase I has demonstrated a complete spectrum of image processing algorithms formulated in log-polar coordinates, proving the feasibility and efficiency of such a system. Orders-of magnitude reduction in computational requirements suggest dramatic benefits for robot vision in general and mobile robots in particular. The proposed technology could improve the cost/performance ration for robot vision systems by several order of magnitude. This opens commercial niches unattainable by conventional Cartesian coordinate based computer vision. Markets include autonomous floor cleaners, transporters, security robots, and service robots.