摘要
传统的GPU实时图像渲染跟踪算法采用多源节点盲分离调度分配方法,整体的执行效率低,图像渲染效果不好.提出一种基于目标分布场相似性度量的实时图形跟踪渲染算法.通过对图像自然分层,保留原始图像的基本信息,在分布场目标跟踪算法中,在图像域进行匹配,采用梯度下降法检测目标,在最优位置处,目标分布场与候选分布场域的相似度能够获得最大响应,由此实现了目标分布场相似度度量,提高对实时图形的跟踪渲染能力.仿真结果表明,采用该算法进行实时图形渲染,可以提高渲染跟踪效率,而且相似度越低,克服多示例学习跟踪和分布场跟踪速度较慢,易陷入局部极小值的不足,性能优越传统方法.
GPU real-time image rendering traditional tracking algorithm using blind separation of multi-source node scheduling and allocation method, the overall low efficiency, image rendering effect is not good. Put forward a similar target distribution field real-time graphics rendering algorithm based on the tracking measurement. Based on the natural hierarchical image, retaining the basic information of the original image, the field distribution in the object tracking algorithm, matching is performed in the image domain, target detection using gradient descent method, in the optimal position, field and the candidate distribution field similarity can obtain the maximum response distribution of the object, thus to realize the goal of field distribution similarity measure, to improve the tracking ability of real-time graphics rendering. The simulation results show that, using the algorithm of real-time graphics rendering, can improve the efficiency of rendering tracking, and the similarity is low, to overcome the multiple instance learning track and field distribution of insufficient tracking slow, easy to fall into local minimum, the superior performance of traditional method.
出处
《微电子学与计算机》
CSCD
北大核心
2015年第7期95-98,102,共5页
Microelectronics & Computer
基金
河南省软科学研究计划项目(132400410934
132400410927)
关键词
GPU
目标分布场
图像
渲染
GPU
target distribution field
image
rendering
