Image Stabilization Using Classified Unit Areas for Motion Correction

The image is divided into multiple unit areas, and local motion vectors are extracted from each unit area. By segmenting the image into manageable regions, the system can process multiple local motion vectors simultaneously, increasing the total number of vectors available for calculating the global motion vector without proportionally increasing computational complexity. This segmentation allows the system to overcome the limitation of using only a limited number of LMVs while maintaining real-time processing capability.

The patent applies different processing strategies to different unit areas based on their characteristics. Unit areas are classified into background, foreground, and short-term motion areas, and valid feature points are selectively extracted based on this classification. This local quality approach ensures that only reliable local motion vectors from background areas are used for global motion calculation, filtering out contaminated vectors from dynamic objects while maintaining high processing efficiency.

The patent performs preliminary classification of unit areas into background, foreground, and short-term motion areas before extracting local motion vectors. This preliminary action allows the system to pre-identify which unit areas will contribute valid motion vectors and which will be excluded. By performing this classification beforehand, the system avoids the computational waste of extracting and then discarding invalid vectors from dynamic objects, thus reducing overall computational load while ensuring high accuracy from the remaining valid vectors.

The system dynamically adjusts the range of the background model based on motion data to optimize the balance between accuracy and computational load. When motion characteristics indicate stable conditions, the model range can be expanded to include more unit areas for higher accuracy. When dynamic objects are detected or motion increases, the range is reduced to exclude contaminated areas. This parameter adjustment allows the system to adaptively control the number of local motion vectors processed, maintaining high accuracy when possible while reducing computational load when necessary.

The patent implements a dynamic background model whose range is adjusted based on motion data from the input frame. The system continuously monitors motion characteristics and adapts the model range accordingly. When the scene is stable, the background model covers a larger portion of the image for better classification accuracy. When dynamic objects appear or motion increases, the model range is reduced to maintain reliability. This dynamic adjustment allows the system to optimize the trade-off between precision and reliability in real-time according to scene conditions.

The system uses motion data extracted from the input frame as feedback to adjust the background model range. The motion characteristics serve as feedback signals that indicate whether the current model range is appropriate. If motion analysis detects large dynamic objects or abnormal motion patterns, the feedback triggers a reduction in model range to exclude contaminated areas. This feedback mechanism ensures the background model maintains high reliability by continuously adapting to changing scene conditions while preserving measurement precision when conditions are favorable.