In this paper, a fast background subtraction algorithm for freely moving cameras is presented. A nonparametric sample consensus model is employed as the appearance background model. The as-similar-as-possible warping technique, which obtains multiple homographies for different regions of the frame, is introduced to robustly estimate and compensate the camera motion between the consecutive frames. Unlike previous methods, our algorithm does not need any preprocess step for computing the dense optical flow or point trajectories. Instead, a superpixel-based seeded region growing scheme is proposed to extend the motion cue based on the sparse optical flow to the entire image. Then, a superpixel-based temporal coherent Markov random field optimization framework is built on the raw segmentations from the background model and the motion cue, and the final background/foreground labels are obtained using the graph-cut algorithm. Extensive experimental evaluations show that our algorithm achieves satisfactory accuracy, while being much faster than the state-of-the-art competing methods.
With the development of virtual reality, there is a growing demand for 3D modeling of real scenes. This paper proposes a novel 3D scene reconstruction framework based on multi-aperture images. Our framework consists of four parts. Firstly, images with different apertures are captured via programmable aperture. Secondly, we use SIFT method for feature point matching. Then we exploit binocular stereo vision to calculate camera parameters and 3D positions of matching points, forming a sparse 3D scene model. Finally, we apply patch-based multi-view stereo to obtain a dense 3D scene model. Experimental results show that our method is practical and effective to reconstruct dense 3D scene.
This paper proposes a fast and accurate algorithm for indirect illumination. It uses volumes of different resolutions to sample and cache the geometric information and the secondary lights. By dividing the irradiance into two parts, it treats the lights coming from the far-field and that coming from the near-field differently. For the far-field ones, it propagates sphere harmonic represented lights on coarse voxels. For the near-field ones, it shoots rays and collects their contributions on fine voxels. By doing this, the algorithm in this paper avoids using many rays to march long distance. In the experiments, it renders about ten times faster than the VGI algorithm to get the same image qualities, especially for the large and complex scenes. Meanwhile, it further accelerates the rendering by inventing an incremental multi-resolution gathering. The experiments illustrate fast and accurate indirect light effects.
Disparity estimation is a popular and important topic in computer vision and robotics. Stereo vision is commonly done to complete the task, but most existing methods fail in textureless regions and utilize numerical methods to interpolate into these regions. Monocular features are usually ignored, which may contain helpful depth information. We proposed a novel method combining monocular and stereo cues to compute dense disparities from a pair of images. The whole image regions are categorized into reliable regions (textured and unoccluded) and unreliable regions (textureless or occluded). Stable and accurate disparities can be gained at reliable regions. Then for unreliable regions, we utilize k-means to find the most similar reliable regions in terms of monocular cues. Our method is simple and effective. Experiments show that our method can generate a more accurate disparity map than existing methods from images with large textureless regions, e.g. snow, icebergs.