Autonomous Mobile Vehicle by Vision (an application of data-flow processor for vision system)

本研究开发了一个同时使用数据流处理器(PD7281：一种反冯诺依曼处理器)和一个冯诺依曼处理器(I80186)的协作式CPU系统。该系统用视觉控制车辆。数据流处理器的缺点是：1)程序开发环境不够好；2)处理器不适合顺序算法。为了改善这些弊端，构建了合作机制。系统的任务分担表现在：1)冯-诺依曼处理器进行控制、判断和识别。这些任务是用通常的顺序算法来描述的。2)数据流处理器进行过滤，而低级图像处理这些任务需要巨大的处理能力。通过使用数据流处理器，获得了以下性能1)120[毫秒]用于帧中的障碍物识别2)13[毫秒]用于帧中的线跟踪。这一性能足以满足车辆视觉导航的需要。在含有障碍物的环境中进行了直线跟踪。结果表明，该系统是有效的。

A collaborationg CPU system which uses both eitht data-flow processers ( PD7281 : one type of anti-von-Neumann processer) and a von-Neumann processer (i80186) had been developed in this research. And the system controlls a vehicle with vision. The results show the effectiveness of the CPU system.Disadvantages of data-flow processers are:1) Environment for development of program is not enough.2) The processers are not suitable for sequential algorithm. In order to improve the disadvantages the collaborating system has been constructed. The share of tasks in the system is shown in the following:1) A von-Neumann processer conducts the control, judgement, and recognition. These tasks are described in usual sequential algorithm.2) Data-flow processers conduct filtering, and low-level image processing These tasks require enormous amount of processing power. By use of data-flow processers following performance was obtained.1) 120[msec] for obstacle recognition in a frame.2) 13[msec] for line trace in a frame. The performance was enough for vehicle guidance by vision. Line trace in the environments which contain obstacles has been carried out. The results show the effectiveness of the system.