Real-Time Moving Object Detection from Airborne Videos with Relative Distance Change Measure

摘要

In this paper, we propose a novel approach for moving object detection from airborne videos. Firstly, features are extracted and tracked between consecutive frames to get matched feature pairs. Then, a novel measure, called relative distance change (RDC), is proposed to classify the matched feature pairs into moving ones and background ones. The proposed RDC measure is invariant to image translation, rotation, and scaling without implementing stabilization of the frames, thus is robust and fast. After that, the moving features are clustered to get candidate moving objects using a growing strategy. Finally, temporal consistence information is utilized to validate and refine the detection results. The proposed method is very fast since the subsequent procedures are performed in the feature pair domain, not in the pixel-by-pixel image domain. Experimental results on four airborne videos show that our method can achieve satisfactory precisions and outstanding real-time performance.