【作者】 付鹰；

【导师】 蔡自兴；

【作者基本信息】 中南大学 ， 计算机软件与理论， 2005， 硕士

【摘要】 环境感知是移动机器人导航研究的关键技术之一。在机器人导航领域中,视觉导航具有分辨率高、探测范围广等优点,缺点是图像计算量大。对于高分辨率图像,仅用车载计算机处理,将不能满足机器人导航实时性要求。为此,本文设计了图像处理网格系统,对部分耗时的图像处理任务进行并行处理,提高图像的处理速度。 为了在网格系统上实现图像并行处理功能,本文进行了如下研究。首先,在图像并行处理应用中,大部分并行算法属于迭代同步的数据并行算法,所以本文对协作任务调度进行了深入分析,提出了协作任务预约调度模型,该算法能够有效地缩短图像处理时间,并且还能提高系统的资源利用率;其次,网格系统属于非专用型工作站网络,工作站属主可能向节点提交计算密集型任务,与并行任务竞争机器资源。在这种情况下,为了使任务能够顺利、及时完成,本文设计了基于数据迁移机制的非抢先动态负载平衡算法,该算法能够保证并行任务的执行效率;最后,本文设计了图像处理网格系统,将上述两个算法应用于网格系统的任务调度器。 图像处理网格系统提供任务调度的基本服务,将图像处理算法封装成网格服务,供系统用户调用。系统以最小化任务完成时间和提高系统资源利用率为目标,可以缩短机器人导航任务的执行时间,增加系统处理导航任务的吞吐量;与此同时,本系统还提供了数据迁移功能,可以保证导航任务正确、及时完成。因此,图像处理网格系统对于提高图像的处理速度,满足移动机器人视觉导航的实时性要求具有一定的研究价值。更多还原

【Abstract】 Environment sensing is one of the focus researches for mobile robot vision-navigation. In the field of robot navigation, vision-navigation has some advantages, such as high resolution, wide detection range, but a drawback of large computation costs for image processing. The real-time requirement of the robot can’t be satisfied, if high resolution images are processed only by the onboard computer. So this thesis designs a grid system for image processing to process some time-consuming tasks, which can increase the speed of image processing.To implement the functionality of parallel image processing in the system, some study has been done. Firstly, in parallel image processing applications, major algorithms belong to data parallel iterative synchronous algorithms. So some co-scheduling algorithms are analyzed in detail, and a co-scheduling algorithm with reservations is proposed. This algorithm can shorten image processing time and improve resource utilization. Secondly grid belongs to non-specific NOWs and the workstation owner may submit to the machine a computation intensive task which competes with a parallel task for the resource. Under this situation, to accomplish tasks in time and successfully, a non-preemptive dynamic load balancing algorithm based on data migration is designed. This algorithm can ensure that parallel tasks shall be executed efficiently. Finally, this thesis designs a grid system for image processing whose scheduler employs the two algorithms mentioned above.The grid system for image processing provides some basic scheduling services, and deploys image processing algorithms as grid services which can be used by the system users. The system aims to minimize task response time and for high resource utilization, hence execution time of robot navigation tasks can be decreased, and high system throughput is provided. At last, the system provides the functionality of data migration, so navigation tasks can be accomplished in time and successfully. Thereby the grid system for image processing has some research value for improving the speed of image processing and satisfying the real-time requirement of robot navigation.更多还原