Video Stabilization via 2D Spatiotemporal Trajectory Analysis

The patent segments the video sequence into 2-D spatiotemporal slices at different orientations, allowing independent analysis of motion patterns at different scales and orientations. This segmentation enables parallel processing of different slices, reducing overall computational complexity while maintaining comprehensive motion detection capability.

The patent transforms the traditional 3-D spatiotemporal volume analysis into multiple 2-D slice analyses. By converting the problem from analyzing the entire 3-D volume to analyzing multiple 2-D slices at different orientations, the computational complexity is reduced while still capturing comprehensive motion information through the multi-orientation approach.

Instead of applying computationally intensive optical flow estimation to all pixels, the patent applies motion analysis only to detected trajectory points in the 2-D spatiotemporal slices. This partial action approach maintains motion estimation precision for relevant features while significantly reducing the computational energy required compared to full-field optical flow.

The patent replaces the traditional optical flow mechanical computation with a trajectory-based analysis method. By detecting trajectories through spatial-temporal correlation in 2-D slices and analyzing their orientations, the system achieves comparable motion estimation precision with reduced computational energy requirements.

The patent segments the 3-D spatiotemporal volume into multiple 2-D slices at different orientations (horizontal, vertical, and oblique). Each slice captures motion information from different perspectives, and the combination of all slices provides comprehensive motion analysis. This segmentation reduces memory requirements and processing complexity compared to analyzing the full 3-D volume at once.

The patent approaches the 3-D motion analysis problem by analyzing multiple 2-D slices at different orientations rather than processing the entire 3-D volume directly. This dimensionality reduction strategy maintains comprehensive motion information coverage while significantly reducing the computational and memory complexity of the processing system.

The patent moves the analysis from traditional spatial domain frame-by-frame comparison to 2-D spatiotemporal slices that incorporate time as an explicit dimension. By analyzing trajectories through space-time correlation in these slices, the method can distinguish between background motion (including camera motion) and foreground object motion more reliably, maintaining detection simplicity while improving reliability in dynamic settings.

The patent introduces 2-D spatiotemporal slices as an intermediary representation between the raw video frames and the final motion detection results. These slices serve as a mediator that captures both spatial and temporal information, enabling more reliable distinction between background and foreground motion while keeping the overall detection process simple and efficient.