An Optical Flow based Moving Objects Detection Algorithm for the UAV

Study on moving objects detection from the UAV's camera has been increasingly emphasized with wide application of the UAV. It is a challenging problem to detect moving objects from moving background due to the motion of the camera. This paper proposes a novel moving objects detection algorithm aimed at the complex changed background in image sequences captured by the UAV's camera. The algorithm distinguishes objects from background by the inconsistency of optical flow, which adopts remapping error of points through homography transformation to extract motion regions firstly. Furthermore, a cluster and convex hull based foreground refinement strategy is proposed to ensure the integrity of detected objects. To deal with large area noise, a false foreground discriminant criterion based on spatiotemporal consistency is designed in this paper. In addition, a frame skipping strategy is proposed to adjust the detection interval based on optical flow vector size for accelerating our algorithm. Extensive experiments show our algorithm achieves outstanding detection performance on VIVID benchmarking dataset, including 5 challenging image sequences recorded in UAV's cameras.