Object Tracking Based on Deep CNN Feature and Color Feature

In this paper, in order to improve the tracking performance of object tracking, the interesting object is modeled by its deep convolutional neural network feature (deep CNN feature) and its color histogram feature. Considering the information of the interesting objects cannot be obtained in advance in actual application, the deep CNN features of the interesting objects are abstracted by the well pre-trained model-VGG-Face. And then the deep feature is combined with color histogram in an adaptive mode to model the object. Finally, an adaptive particle filter algorithm based on deep CNN feature and color feature is proposed to track the interesting object. The experimental results show that the proposed method can deal with serious object occlusions and appearance changes.