Meandering Scanning Order for Video Data Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current data processing systems for video data face challenges in reducing chip area requirements without impairing processing performance or quality, particularly in video processing where motion estimation involves large search areas that require significant memory resources and power consumption.
Innovation Solution
A data processing system utilizing a two-level memory hierarchy with a meandering scanning order orthogonal to the main direction, allowing continuous data fetching and processing without breaking the scanning order, and enabling efficient parallel processing by shifting data on a line-wise basis between L1 and L0 buffers.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Area of stationary object
If a two-level memory hierarchy with L0 and L1 buffers is used to reduce chip area, then chip area is reduced, but scanning interruptions occur which impair processing continuity
Solution Approach 1:
The patent applies dimensionality change by switching from a conventional linear scanning order to a meandering (snake-like) scanning pattern that traverses the search area in an orthogonal direction. This allows the scanning to continue uninterrupted across buffer boundaries by changing the scanning dimension, thereby maintaining processing continuity while using smaller L1 buffers.
Solution Approach 2:
The search area is divided into multiple blocks that can be processed independently. The meandering scanning order processes these blocks in a systematic sequence that ensures continuous data flow between L0 and L1 buffers without requiring the entire search area to be loaded at once, thus reducing chip area while maintaining reliability.
2Area of stationary object
If the L1 buffer size is reduced to save chip area, then chip area is reduced, but data fetching efficiency deteriorates
Solution Approach 1:
By changing the scanning dimension to a meandering pattern, the system can efficiently utilize a smaller L1 buffer. The orthogonal traversal allows data to be fetched in a compact sequence that fits within reduced buffer capacity while maintaining high data fetching efficiency through continuous memory access patterns.
Solution Approach 2:
The meandering scanning order pre-establishes a systematic data fetching pattern that optimizes memory access before actual processing begins. This preliminary organization of data access allows the smaller L1 buffer to be efficiently utilized, maintaining data fetching efficiency without requiring large buffer capacity.
3Device complexity
If a conventional linear scanning order is used, then data fetching is simple, but scanning interruptions occur at buffer boundaries
Solution Approach 1:
The patent resolves the contradiction by introducing a meandering scanning pattern that changes the scanning dimension. This orthogonal traversal method eliminates buffer boundary interruptions by systematically moving through blocks in a continuous path, maintaining scanning continuity while adding manageable complexity to the data fetching process.
4Measurement precision
If the search area is enlarged to improve motion estimation accuracy, then picture quality is improved, but memory bandwidth requirements increase
Solution Approach 1:
The patent segments the enlarged search area into multiple smaller blocks that are processed using meandering scanning. This segmentation allows the system to handle large search areas by processing them in manageable chunks, reducing the peak memory bandwidth requirements while maintaining the overall accuracy benefits of the enlarged search area.
Solution Approach 2:
By applying meandering scanning in an orthogonal direction, the system can efficiently process enlarged search areas with reduced memory bandwidth requirements. The dimensional change in scanning pattern allows for more efficient memory utilization and data fetching, accommodating larger search areas without proportionally increasing bandwidth demands.
Data Source
AI summary
A data processing system is provided for processing video data on a window basis. At least one memory unit (L1) is provided for fetching and storing video data from an image memory (IM) according to a first window (R) in a first scanning order. At least one second memory unit (L0) is provided for fetching and storing video data from the first memory unit (L1) according to a second window in a second scanning order (SO). Furthermore, at least one processing unit (PU) is provided for performing video processing on the video data of the second window as stored in the at least one second memory unit (L0) based on the second scanning order (SO). The second scanning order (SO) is a meandering scanning order being orthogonal to the first scanning order (SO1).