Compact representations of videos through dominant and multiple motion estimation

Sawhney, Harpreet; Ayer, Steven K.

doi:10.1109/34.531801

Cited by 257 publications

(143 citation statements)

References 32 publications

Supporting

Mentioning

143

Contrasting

Order By: Relevance

“…Particularly, Wang and Adelson [1] estimate object motions in successive frames and track them through the sequence by computing optical flow vectors, fit affine motion models to these vectors, and then cluster the motion parameters into a number of objects using k-means. Darrell and Pentland [8], and Sawhney and Ayer [9] used similar approaches based on optical flow estimation between successive frames and apply the MDL principle for selecting the number of objects. Note that a major limitation of optical-flow based methods concerns regions of low texture where flow information can be sparse, and when there is large inter-frame motion.…”

Section: Related Workmentioning

confidence: 99%

Sequential Learning of Layered Models from Video

Titsias

Williams

2006

Toward Category-Level Object Recognition

View full text Add to dashboard Cite

Abstract. A popular framework for the interpretation of image sequences is the layers or sprite model, see e.g. [1], [2]. Jojic and Frey [3] provide a generative probabilistic model framework for this task, but their algorithm is slow as it needs to search over discretized transformations (e.g. translations, or affines) for each layer simultaneously. Exact computation with this model scales exponentially with the number of objects, so Jojic and Frey used an approximate variational algorithm to speed up inference. Williams and Titsias [4] proposed an alternative sequential algorithm for the extraction of objects one at a time using a robust statistical method, thus avoiding the combinatorial explosion. In this chapter we elaborate on our sequential algorithm in the following ways: Firstly, we describe a method to speed up the computation of the transformations based on approximate tracking of the multiple objects in the scene. Secondly, for sequences where the motion of an object is large so that different views (or aspects) of the object are visible at different times in the sequence, we learn appearance models of the different aspects. We demonstrate our method on four video sequences, including a sequence where we learn articulated parts of a human body.

show abstract

Section: Related Workmentioning

confidence: 99%

Sequential Learning of Layered Models from Video

Titsias

Williams

2006

Toward Category-Level Object Recognition

View full text Add to dashboard Cite

show abstract

“…This technique has been successfully applied to many different applications like video compression [1], video indexing [2], or creation of virtual environments [3]. For example, Shum and Szeliski [3] proposed a method to stitch a set of images together for constructing a panorama.…”

Section: Introductionmentioning

confidence: 99%

Fast stitching algorithm for moving object detection and mosaic construction

Hsieh

2003

2003 International Conference on Multimedia and Expo. ICME '03. Proceedings (Cat. No.03TH8698)

View full text Add to dashboard Cite

This paper proposes a novel edge-based stitching method to detect moving objects and construct mosaics from images. The method is a coarse-to -fine scheme which first estimates a good initialization of camera parameters with two complementary methods and then refines the solution through an optimization process. The two complementary methods are the edge alignment and correspondence-based approaches, respectively. Since these two methods are complementary to eac h other, the desired initial estimate can be obtained more robustly. After that, a Monte-Carlo style method is then proposed for integrating these two methods together. Then, an optimization process is applied to refine the above initial parameters. Since the found initialization is very close to the exact solution and only errors on feature positions are considered for minimization, the optimization process can be very quickly achieved. Experimental results are provided to verify the superiority of the proposed method. IntroductionImage stitching is the process of recovering the existing camera motion parameters between images and then compositing them together. This technique has been successfully applied to many different applications like video compression [1], video indexing [2], or creation of virtual environments [3]. For example, Shum and Szeliski [3] proposed a method to stitch a set of images together for constructing a panorama. In addition, Irani and Anandan [2] used this technique to represent and index different video contents. For most methods in this field, an affine camera model is commonly used to approximate possible motions between two consecutive frames. Then, this model can be recovered by two common methods, i.e., the correlation-based approach and the optimization-based one. For example, Kuglin and Hines [4] presented a phase-correlation method to estimate the displacement between two adjacent images in frequency domain. In addition, Zoghlami et al. [5] proposed a corner-based approach to build a set of correspondences for computing possible transformation parameters from pair of images.However, the establishment of good correspondences is a challenging work when images have nonlinear intens ity changes [3]. In order to avoid this problem, Szeliski [3] proposed a nonlinear minimization algorithm for automatically registering images by minimizing the discrepancy in intensities between images. In comparison with the correlation-based method, the global optimization approach performs more robustly but will be trapped on a local minimum if the starting point is not properly initialized.In this paper, we present an edge-based stitching technique to detect moving objects and construct mosaics from consecutive images. In general, the motion model between consecutive images is non-linear. This paper uses a coarse-to-fine approach to robustly and accurately recover this nonlinear model. I n the coarse stage, two complementary methods, i.e., the edge alignment and the correspondence-based approaches, are first proposed t o get respe...

show abstract

“…The more recent interest on the so-called layered representations for video [4,5,6,7,8] has motivated further work on motion-based segmentation. A number of approaches in the computer vision literature uses other cues besides motion, such as color and edges [9], or regularization priors [10].…”

Section: Introductionmentioning

confidence: 99%

Occlusion-Based Accurate Silhouettes from Video Streams

Aguiar

Miranda

Castro

2006

Lecture Notes in Computer Science

View full text Add to dashboard Cite

Abstract.We address the problem of segmenting out moving objects from video. The majority of current approaches use only the image motion between two consecutive frames and fail to capture regions with low spatial gradient, i.e., low textured regions. To overcome this limitation, we model explicitly: i) the occlusion of the background by the moving object and ii) the rigidity of the moving object across a set of frames. The segmentation of the moving object is accomplished by computing the Maximum Likelihood (ML) estimate of its silhouette from the set of video frames. To minimize the ML cost function, we developed a greedy algorithm that updates the object silhouette, converging in few iterations. Our experiments with synthetic and real videos illustrate the accuracy of our segmentation algorithm.

show abstract

Compact representations of videos through dominant and multiple motion estimation

Cited by 257 publications

References 32 publications

Sequential Learning of Layered Models from Video

Sequential Learning of Layered Models from Video

Fast stitching algorithm for moving object detection and mosaic construction

Occlusion-Based Accurate Silhouettes from Video Streams

Contact Info

Product

Resources

About