Spatial Multilevel Optical Flow Architecture for Motion Estimation of Stationary Objects with Moving Camera

공간 다중레벨 Optical Flow 구조를 사용한 이동 카메라에 인식된 고정물체의 움직임 추정

This paper introduces an approach to detect motion areas of stationary objects when the camera slightly moves in the scene by computing optical flow. The flow field is computed by two pyramidal architectures of 5 levels which are built by down-sampling the size of the images by half at each level. Two pyramids of images are built and then optical flow is computed at each level. A warping process combines the information and generates a final flow field after applying edge smoothness and outliers reduction steps. Moreover, we convert the flow vectors in order of magnitude and angle to a color map using a pseudo-color palette. Experimental results in the Middlebury optical flow dataset demonstrate the effectiveness of our method compared to other approaches.