Robust Global Motion Estimation Oriented to Video Object Segmentation

Abstract

Most global motion estimation (GME) methods are oriented to video coding while video object segmentation methods either assume no global motion (GM) or directly adopt a coding-oriented method to compensate for GM. This paper proposes a hierarchical differential GME method oriented to video object segmentation. A scheme which combines three-step search and motion parameters prediction is proposed for initial estimation to increase efficiency. A robust estimator that uses object information to reject outliers introduced by local motion is also proposed. For the first frame, when the object information is unavailable, a robust estimator is proposed which rejects outliers by examining their distribution in local neighborhoods of the error between the current and the motion-compensated previous frame. Subjective and objective results show that the proposed method is more robust, more oriented to video object segmentation, and faster than the referenced methods.

Robust camera motion estimation in presence of large moving objects

“…We compare our algorithm with the robust method proposed in [8] to estimate the camera motion in the first frame of a sequence. This "reference method" uses a hierarchical differential-based scheme and a robust estimator to ignore outliers from moving objects by exploiting their spatial coherence.…”

“…In this paper we also follow this convention. GME algorithms can derive the camera motion either from the image intensity values (as phase correlation [6] [10] and differential based techniques [5] [8]) or from some previous estimation of local motion (as feature based [3] [9] and optical flow based techniques [7] [11]). But their suitability for a particular application mostly depends on their behavior with data contaminated by pseudo-outliers (data from alternative structures, i.e.…”

“…Since objects can so severely jeopardize GME, some approaches like [8] devote an initial stage to discard them beforehand. However, this technique relies on thresholding the residuals of a non-robust initial model, which makes it acceptable only for small objects, which do not distort too much that model.…”

Hierarchical block-based image registration for computing multiple image motions

“…The work presented in [7] is somewhat in between. A block based approach is used for the object detection stage but not for the global motion estimation, and the object detection uses a difference based approach rather than using the information available from the registration.…”

Research on Motion Attention Fusion Model-Based Video Target Detection and Extraction of Global Motion Scene

“…The difficulty in moving target detection and extraction under global motion scene is that video motion characteristic is the result of the superposition of global motion and local motion. The most effective solution at present is the detection algorithms based on motion compensation [9][10]. Its main clue is to use six-parameter affine model to estimate global motion, then recursive least square is adopted to calculate model parameters, obtain the relative motion between moving target and background utilizing motion compensation and finally acquire target region (TR).…”