Multi-Features Integration Based Hyperspectral Videos Tracker
Zhe Zhang, Kun Qian, Juan Du, Huixin Zhou
Abstract
Most target tracking is over visible videos, but in a challenging scene, tracking targets with the same appearance is very difficult on visible videos, due to the limitation of grayscale and color information. Therefore, we use Hyperspectral Videos (HSVs) with rich spectral information for target tracking to distinguish similar targets. In this paper, a multi-features integration based tracking method is proposed over HSV. The feature maps are generated by Histogram of Gradient (HOG) and pretrained VGG-19 network, and then kernelized correlation filter framework is utilized to detect target over HSVs. Specially, More information of spatial, spectral and temporal are all used to extract useful features, and these feature can track the target that can not be tracked in visible videos. The experimental results on HSVs show that the proposed method has better performance than the three existing tracking methods with hyperspectral information.