Stereoscopic Scene Flow Computation for 3D Motion Understanding

摘要: Building upon recent developments in optical flow and stereo matching estimation, we propose a variational framework for the estimation of stereoscopic scene flow, i.e., motion points three-dimensional world from image sequences. The proposed algorithm takes into account pairs two consecutive times computes both depth 3D vector associated with each point image. In contrast to previous works, partially decouple which has many practical advantages. formulation is quite flexible can handle sparse or dense disparity maps. method very efficient; map being computed on an FPGA, GPU, runs at frame rates 20 frames per second QVGA images (320×240 pixels). Furthermore, present solutions important problems estimation: violations intensity consistency between input images, uncertainty measures result.