Hyperspectral Video Tracker Based on Anomaly Suppression and Multi-Feature Integration
Huihui Guo, Xu Yang, Zebin Wu, Zhihui Wei · 2023
Due to the rich spectral information provided by hyperspectral images(HSIs), hyperspectral trackers are better suited for complex scenes compared to color image trackers. This paper proposes a tracking method suppressing anomalies during the detection process, addressing issues like occlusion and illumination variation that cause changes in the target’s appearance, thereby enhancing the performance of the hyperspectral target tracker. Simultaneously, a pre-trained convolutional network and Histogram of Oriented Gradient(HOG), along with abundance information, are leveraged to generate multiple features. These different features are integrated using adaptive weights. Then, a Background-Aware Correlation Filter(BACF) tracker is employed to detect targets. Experimental results demonstrate that this method achieves promising performance on hyperspectral videos and outperforms three other advanced tracking methods in the same scene.