改进的卷积网络目标跟踪算法

在目标跟踪算法中深度网络可以对大量图像进行训练和表示，但是对于特定的跟踪对象，离线训练不仅费时，而且在对大量图像进行学习时，其表示和识别能力效果不佳。基于以上问题提出有模板更新的卷积网络跟踪算法，可以在没有离线训练的大量数据时，也能够利用实现强大的目标跟踪能力。在目标跟踪中，从目标周围区域提取一组归一化的局部小区域块作为新的滤波器，围绕目标定义下一帧中的一组特征映射来提取自适应滤波器周围目标，对随后帧提取的归一化样本进行卷积操作生成一组特征图；利用这些特征图获取每个滤波器和目标的局部强度衍射图样之间的相似性，然后对其局部结构信息进行编码；最后，使用来自全局表示的特征图保存该目标的内部几何设计，再通过软收缩方法去噪抑制噪声值，使其低于自适应阈值，生成目标的稀疏表示。有模板更新改进的CNT算法能稳定地跟踪目标，不会发生严重漂移，具有优于传统CNT的良好跟踪效果。

Improved convolutional network target tracking algorithm

Deep networks have been successfully applied to visual tracking by training large numbers of images offline. However, the offline training is time-consuming and the learned vast representation may be less discriminative for tracking specific objects. In this article, although without training deal auxiliary data, simple two-layer convolutional networks can be powerful enough to learn robust representation for visual tracking. In the first frame, this algorithm extracted a set of normalized patches from the target region as fixed filters, which integrated a series of adaptive filters surrounding the target to define a set of feature maps by extracting the normalized samples of the subsequent frames of convolution operation. Furthermore, all the maps together form a global representation, via which the inner geometric layout of the target was also preserved. A simple soft shrinkage method that suppresses noisy values below an adaptive threshold is employed to de-noise the global representation. The convolution networks have a lightweight structure that is good for methods on the recent tracking benchmark data set with multiple videos.