AV1 Zeromv in Motion Vector Search Hierarchy

The ZEROMV candidate in the AV1 video codec is a baseline motion vector representing zero spatial displacement (0, 0) between the current block and its reference frame. Within AV1's motion vector prediction hierarchy, it functions as a primary predictor for static visual elements and serves as a default fallback to ensure candidate lists remain fully populated. This article examines the definition of ZEROMV, its role within the AV1 Dynamic Reference List (DRL), and how it optimizes coding efficiency for stationary content.

Definition of the ZEROMV Candidate

In inter-frame prediction, motion vectors indicate the displacement of a block relative to a previously decoded reference frame. The ZEROMV candidate represents a null vector where both horizontal and vertical displacements equal zero (dx = 0, dy = 0).

When a block selects ZEROMV, the encoder copies the sample values directly from the identical coordinate position in the specified reference frame. Unlike GLOBALMV, which derives motion from calculated camera parameters (such as pans, tilts, or zooms), ZEROMV strictly denotes zero translational displacement without applying affine or perspective transformations.

Placement in the Motion Vector Search Hierarchy

AV1 constructs a candidate list of motion vectors from adjacent spatial neighbors (left, above, and surrounding corners) and temporal collocated blocks. These candidates are ranked and prioritized through a structured hierarchy:

1. Spatial and Temporal Predictors: The encoder first evaluates NEARMV and NEWMV candidates derived directly from neighboring blocks.
2. Global Predictors: AV1 checks for GLOBALMV, which incorporates frame-level global motion estimation.
3. ZEROMV Insertion: If neighboring blocks do not provide enough distinct candidates to fill the candidate list, or if the spatial neighbors themselves possess zero displacement, ZEROMV is introduced into the list.

ZEROMV acts as an anchor in this hierarchy. AV1 mandates that the candidate list must maintain a defined length for entropy coding consistency. When unique spatial or temporal predictors are exhausted, the encoder pads the remaining slots with ZEROMV.

Compound Prediction Modes

AV1 supports compound prediction, where two reference frames are combined to predict a single block. ZEROMV integrates into several compound motion modes, including:

• NEAR_ZEROMV: Combines a near spatial predictor for the first reference frame with a zero vector for the second reference frame.
• ZEROMV_NEAR: Uses a zero vector for the first reference and a near predictor for the second.
• ZEROMV_ZEROMV: Applies a (0, 0) vector across both reference frames, blending two temporally distinct instances of the same spatial location.

Coding Efficiency and Performance

The ZEROMV candidate minimizes bit overhead. Because zero-motion states are statistically common in video—such as stationary backgrounds, locked-off camera shots, and non-moving UI elements—AV1’s entropy coder assigns high probability and minimal bit cost to ZEROMV signaling. When selected, the encoder transmits minimal syntax data because no motion vector difference (MVD) values need to be coded, directly reducing the overall bitrate while maintaining visual fidelity.