Underwater Vision For Intelligent Autonomous Underwater Vehicles

佛罗里达大西洋大学和卡内基-梅隆大学提出了一项联合工作计划，以开发用于智能自主水下机器人的水下三维视觉系统。这项工作的主要重点将是开发用于处理来自多单元声纳阵列的信号的算法，以生成海底地形的网格表示，并开发地图构建算法，以生成连贯的地形地图，并将该表示用于导航、目标识别和分类以及路径规划。目前类型的声纳系统通常生成二维信息，或者如果生成三维信息，则分辨率相当差。因此，将开发一种创新的声纳系统，它在工作模式上与其他声纳系统不同，更依赖于数据处理，这将允许以高分辨率提取三维信息，并具有仅由处理单元的速度控制的刷新率。信号处理的基本方法是对声纳系统的多单元接收器阵列接收到的声信号在时间和空间上进行多维离散傅立叶变换。该处理将首先生成角度-角度-范围数据，然后将其转换为海底地形的离散表示。地图制作过程的主要任务将是数据处理和从不同车辆位置和相对于地形的方位获得的数据结构的匹配，以及提取物体位置、物体类型和地形信息等高层特征。其结果是支持水下测量和基于传感器的导航所需的高级别三维世界模型图像。

A joint program of work is proposed between Florida Atlantic University and Carnegie-Mellon University to develop an underwater three dimensional vision system for intelligent autonomous underwater vehicles. The main thrust of the work will be in the development of algorithms for the processing of the signal from a multiple element SONAR array to generate a grid representation of the ocean bottom terrain and the development of map building algorithms to generate a coherent map of the terrain and use the representation for navigation, object recognition and classification, and path planning. The present types of SONAR systems in general generate two dimensional information or if three dimensional information is generated, the resolution is rather poor. Thus, an innovative SONAR system is to be developed that differs from other SONAR systems its in its mode of operation, relying more on data processing which will allow the extraction of three dimensional information with high resolution and with a refresh rate controlled only by the speed of the processing unit. The basic approach of the signal processing consists of multi-dimensional discrete Fourier transforms, in time and space, of the acoustic signals received by the multi-element receiver array of the SONAR system. The processing will first generate angle-angle-range data which is then converted to a discrete representation of the ocean bottom terrain. The main tasks of the map building process will be the data handling and the matching of data structures acquired from different vehicle locations and orientations relative to the terrain and the extraction of high level features such as object location, object type an topographical information. The result is a high level three dimensional world model image required to support underwater surveying and sensor based navigation.