基于跟踪度的Gabor小波特征跟踪方法的研究

图像的边缘包含了目标的大量特征信息。利用Gabor小波可以从目标图像中提取具有特征位置、角度和尺度的参数。利用这些参数可以重建除均值以外的所有图像信息。图像边缘与均值无关。根据重建图像边缘线段的长度和边缘拟和度以及特征点的个数,提出了跟踪度的概念,分析了跟踪度的性质,并在跟踪度准则指导下确定了跟踪的特征点个数。仿真实验证明,跟踪度反映了目标特征跟踪的可靠程度,提供了跟踪精度的客观标准,为选择特征点个数、平衡计算复杂度和跟踪精度提供了客观的依据。通过对目标进行姿态变换和大面积遮挡的跟踪实验证明,当跟踪度达到0 95以上时就可以稳定地跟踪目标。

Gabor wavelet feature tracking method based on criteria of tracking

The image edges included most feature information of the target. When the feature with Gabor wavelet was extracted from the target image, the feature's position, angle and scale information could be decided. With these information the image could be reconstructed without average gray. But the edges were independent of the average. According to the length of the edge of reconstruction image and its similitude degree with the source edge and the number of feature, the criteria of tracking (COT) was defined, and its property was analyzed. With this COT how many feature should got during the tracking to get enough precision, and spend less time could be decided. The results of experiment show that the criteria of tracking reflects the reliability of target feature tracking, and provides an impersonality criteria of tracking precision. It gives out a measure to balance the computation complex and tracking precision. The experiments on gesture changing and partly blocked plane show that a target can be reliably tracked when COT is more than 0.95.