摘要
在跟踪-学习-检测(tracking-learning-detection,TLD)框架中,跟踪器采用传统的归一化互相关(Normalized Cross Correlation,NCC)算法完成图像的匹配.该方法具有较强抗噪声能力,但计算量庞大,难以满足实时跟踪的要求.对TLD跟踪器的匹配方法进行了改进,匹配过程由粗匹配和精匹配两步完成.新的两步匹配方法在保持较强抗噪能力的同时,减少了运算量,提高了匹配速度.实验表明,采用新的匹配方法的TLD目标跟踪器可以准确快速地进行特征点匹配,减少了计算时间,并有效地降低了误匹配率.
In tracking-learning-detection, we use conventional normalized cross correlation only to do the match. Although it has some capacities of resisting noisy, it is difficult to meet real-time processing tracking requirements because of the amount of calculation of normalized cross correlation algorithm. In this paper, we will make improvement on the matching method of TLD tracker, and the image matching is divided into two stages of rough matching and fine matching. The new method has high anti-noise capability, and in the meanwhile it reduces computation cost and improves the matching speed. Experimental results show that the algorithm carried out matching feature points quickly and accurately, reducing the computational time, and reducing the false match rate effectively.
出处
《小型微型计算机系统》
CSCD
北大核心
2015年第5期1113-1116,共4页
Journal of Chinese Computer Systems