论文部分内容阅读
基于相关的模板匹配方法在图像跟踪中得到了广泛的应用,但是这种方法的主要缺点是对图像噪声非常敏感,而且由这种方法得出的相关曲面比较平坦.为了有效抑制噪声和局部遮挡对图像的影响,提出用鲁棒统计中的一个值代替相关值作为距离度量.它不是计算两幅图像的对应像素差值, 而是计算两幅图像中差别不大的对应像素对的个数.实验证明这样得到的曲面相对要尖锐得多.当上述的距离度量应用于跟踪过程中时, 模板的合理更新对跟踪非常重要.基于上述的距离度量方法, 提出了一种新的模板更新策略.利用两幅图像中近似像素在整修匹配模板像素中所占的比例大小决定加权系数.这种加权策略可以自适应地根据图像序列的变化选择更新相关模板,同时在一定程度上减小模板漂移的危险.“,”Template matching by means of correlation is common practice in object tracking. However, its sensitivity to deformations of the patterns and the broad unsharp peaks are significant drawbacks. To reduce the influence of the noise and occlusion, a novel distance measurement method is proposed, in which first order statistics of the close points between the template and the image are used as a matching measurement. Instead of computing the distance of the gray value of the two images, it counts the number of the close pixel pairs in the object image and the template. The correlation curve got in this way is proved to be much sharper and the matching precision is improved significantly. When above method is applied to the processing of tracking, the reasonable update of the template will be the key. Based on above similarity measurement method, an adaptive template updating strategy is also presented, which is capable of selectively updating the correlation template in response to the changes in an image sequence and enable the correlation to track target under serious background clutter while decreasing the risk of template drift.