Moving Vehicle Detection in Satellite Video via Background Subtraction and Global–Local Features Fusion Faster R-CNN

Abstract

AbstractWith the successful operation of the Jilin-1 constellation and other staring satellites in recent years, satellite videos have become new resources for real-time tasks such as surveillance. However, it is difficult to detect small objects in satellite videos due to the high resolution of the satellite video, and the low contrast between the objects and background. This paper develops a novel small moving vehicle detection scheme by exploiting the spatial–temporal information of satellite video. The proposed method consists of four stages: (i) pre-processing, to filter, adjust the contrast, and register the satellite video frame by frame; (ii) extracting the candidate moving vehicle region based on the improved ViBe method; (iii) detecting the vehicle both moving and stationary by the global–local features fusion Faster R-CNN; and (iv) merging the detected results from previous stages to propose the final results, accidental miss detection will be recalled by the inter frame compensation module. Experimental results show that the proposed method demonstrates better performance even when compared to some state-of-the-art methods.KeywordsMoving vehicle detectionSatellite videoBackground subtraction