Infrared unmanned aerial vehicle targets detection based on multi-scale filtering and feature fusion

A multi-scale filtering and feature fusion target detection algorithm is proposed based on the large-field infrared search system for the detection of Unmanned Aerial Vehicle (UAV) targets at different scales in low-altitude background. In the initial targets detection, the algorithm transforms the image sequence into different scales, then uses the median and Robinson filters to suppress the background. At the stage of false alarm elimination, it firstly fuses the filter results on the original scale, then extracts the characteristics of the fusion image and analyzes their inter-class relations. Finally, it designs a classifier based on confidence scoring mechanism to achieve real object confirmation and false alarm elimination. The experimental results show that the proposed algorithm can effectively eliminate the false alarm that has similar characteristics to the real targets. It has a good detection effect on the low speed moving UAV targets when the dimension prior information is unknown.