连通域标记并行算法在多核处理器上的设计和实现

为满足全方位视觉运动目标检测跟踪系统中检测和跟踪实时性的要求,采用多核程序设计和并行处理技术,对系统中连通域标记算法,重新进行了设计和实现。为解决由图像分割造成的连通域的分离,应用边界处的游程码,将两个不同任务块中分割处的连通域合并成一个连通域。最后通过实验证明该并行算法比传统算法在多核处理器上运行更快,更好地解决了多核处理器饥饿问题,让多核处理器发挥了更高的效率。

Design and Implementation of Connected Component Labeling Parallel Algorithm with Multi-Core Processor

To meet the real-time requirement of moving objects detection and tracing based on omnidirectional vision,multi-core programming and parallel processing technology are applied to the redesign and realization of the connected component labeling algorithm.The paper introduces run-length encoding into algorithm to solve separation of connected component by image segmentation.Two segmented connected components for different tasks' block are merged into one connected component.Experiments show that the parallel algorithm runs faster than the traditional algorithm in multi-core processor.It has better solved the problem of hunger in multi-core processors and made the multi-core processor more efficient.