Infrared Small Target Detection Algorithm Based On Guided Filter And Human Visual Attention Mechanism

Separating the targets from the background effectively is the key point of infrared moving target detection and tracking technology. The infrared image based on the infrared thermal imaging method is characterized by strong noise, multi-disturbance and blurred boundaries especially when the signal is weak or thermal imaging area is small. As a result, the background and noise have a large percentage of the image,which result in more difficulty of detecting infrared small target. This thesis does a profound research on the highly accurate infrared small target detection technology and proposes improved algorithms.Taking into account the characteristics of the infrared image and the difficulty of small target detection, algorithms in this thesis use the guided filter for preprocessing. The guided filter can maintain the characteristics of the target edge while smoothing the background?enhancing association between backgrounds and reducing noise. Human visual attention mechanism has the feature that it can guided the eyes to focus on the significant region of the mass data. Based on this feature, layered threshold and connected component analysis are used to extract the ROIs(Region of interest) of infrared image. Multi-scale LOG filtering is also adopted to improve the target SNR in the ROI. Then suspicious targets can be obtained through SNR threshold preliminarily. In order to detect the true small targets correctly and filter false targets, this thesis proposes an improved moving pipeline filter algorithm by full use of the prior information of the targets in the pipeline.To validate the effectiveness of the guided filter and improved moving pipeline filter, experiments have been done under different conditions, including multi-target?dim target and complex background. The results of the experiments confirm the satisfactory performance of the proposed algorithms in this thesis. Besides, the algorithm is also experimented with different videos. Detection rate and false alarm rate of the results show that the algorithm has good adaptability.