论文部分内容阅读
直升飞机任务的复杂性,包括接近地面的操作,使飞行员工作负担沉重。为了让驾驶员有时间完成定向任务,辅助传感器和一些导航及控制功能的自动化是非常需要的。来自光电传感器的图象为低飞直升飞机在飞行路径上提供检测物体的秘密方法。被动测距就是处理图象序列,采用的方法基于视觉流计算和递归估计。被动测距算法必须从成象中提取障碍物信息,而成象以每秒5到30或更多帧变化,这取决于直升飞机的速度。我们已脱机用直升飞机的图象实现和试验了被动测距算法。尽管如此,算法的实时数据和计算需求是超出任何现行的微处理机或数字信号处理机的能力。本文叙述了算法的计算需求和使用并行处理技术以满足此需求。讨论了选择并行处理结构中的各式各样的问题,按其对实时处理算法的适用性对四种不同的计算机结构进行评价。根据这一评价,我们断定实时被动测距是一个现实的目标且在短期内就可达到。
The complexity of the helicopter mission, including operations close to the ground, put a heavy burden on pilots. Automated sensors and some navigation and control functions are needed to give the driver time to complete the orientation tasks. The image from the photoelectric sensor is a secret way for the low flying helicopter to provide an object of detection on the flight path. Passive ranging is the processing of image sequences. The method used is based on visual flow computation and recursive estimation. The passive ranging algorithm must extract the obstacle information from the imaging and the imaging varies by 5 to 30 or more frames per second, depending on the speed of the helicopter. We have implemented and tested the passive ranging algorithm with the image of a helicopter offline. However, the real-time data and computational requirements of the algorithm are beyond the capabilities of any current microprocessor or digital signal processor. This article describes the algorithmic computational needs and uses parallel processing techniques to meet this need. The various problems in selecting parallel processing architectures are discussed, and four different computer architectures are evaluated in terms of their applicability to real-time processing algorithms. Based on this assessment, we conclude that real-time passive ranging is a realistic goal and achievable in the short term.