滤光片表面缺陷视觉检测系统多处理器调度优化

Multi-processor scheduling of vision detecting system for filter surface defect

针对滤光片表面缺陷视觉检测系统中在线检测实时性需求对检测速度要求较高,研究一种有效利用可用硬件资源并行处理实时工作提高处理速度的调度优化策略。基于AOE图对滤光片表面缺陷视觉检测系统进行任务级分析,优化事件、活动拓扑关系与任务间冗余的数据相关性、资源相关性,建立并行任务模型;采用关联处理器调度算法(arbitrary processor affinities,APAs)进行并行多处理器调度,指定任务只能被某个处理器集合执行,将期限紧迫、缓存敏感的任务限制在单一处理器,提高资源利用率,改进检测系统实时性。试验结果表明:在尺寸为1.20mm×1.20mm、26×28个滤光片组成滤光片面板上,采用多处理器调度可使检测速度极大提升,采用APAs调度算法后,平均缺陷识别完成时间为常规检测系统时间的36.5%,可以满足在线实时要求,证明应用多处理器调度方法,可以极大提升检测仪器实时性能的有效性。

Online real-time detection is required in vision detecting system for filter surface defect. This paper discusses about a scheduling optimization strategy for utilizing available hardware resources to process real-time work in parallel so as to improve the processing. An AOE graph was used to perform task-level analysis on the abovementioned vision detecting system, optimize the data and resource dependency between events, topological relation of activities and inter-task redundancy, and establish a parallel task model. An arbitrary processor affinities( APAs) was employed for parallel multiprocessor scheduling. The designated task could only be executed by a processor set, i.e., the tasks with imminent deadline and sensitive caches were confined on a single processor to enhance the level of resources utilization and improve the timeliness of detecting system. The results show that, on the panel with 26×28 optical filters( 1.20 mm×1.20 mm),multiprocessor scheduling can greatly increase the detection speed. With the APAs algorithm, the average time in identifying defects is 36.5% that of the routine detection system. It has been proven that the multiprocessor scheduling can largely enhance the effectiveness f real-time performance of testing instruments.