Overlapping Region Division for High-Resolution Depth Data Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for acquiring high-accuracy and high-resolution depth data in object recognition and robotics face inconsistencies at image boundaries due to division processing, leading to increased processing load and decreased speed.
Innovation Solution
An information processing apparatus that divides input depth and image data into overlapping regions, performs inference processing on each region, and combines the results to generate high-resolution depth data, reducing the need for data synchronization and allowing for parallel processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If division processing is performed on input images to reduce processing load, then processing speed is improved, but inconsistency occurs at boundary portions of divided images
Solution Approach 1:
The input image is divided into multiple divided images for parallel processing. Each divided image is processed independently by the inference device to generate depth data, enabling parallel computation and improved processing speed while maintaining the ability to handle large-scale images efficiently.
Solution Approach 2:
Overlap regions are introduced at the boundaries of divided images where additional processing is applied. The combining device performs special handling on overlap regions to ensure consistency, applying different processing quality to boundary areas compared to internal regions. This local quality adjustment resolves the inconsistency issue at boundaries while preserving the benefits of division processing.
2Measurement precision
If second inference processing is performed to correct boundary inconsistency, then depth data accuracy is improved, but processing load increases and processing speed decreases
Solution Approach 1:
Overlap regions are prepared in advance during the division stage, before inference processing occurs. By pre-defining the overlap regions and their handling strategy, the system avoids the need for complex post-processing corrections. This preliminary action reduces the processing load during the inference stage while ensuring accuracy in boundary regions.
Solution Approach 2:
The overlap regions are essentially copied from adjacent divided images and processed together. Instead of performing complex second inference processing, the system uses the copied overlap region data from neighboring divisions and combines them appropriately, reducing computational complexity while maintaining boundary consistency.
3Reliability
If data synchronization is required in neural network intermediate layer, then boundary inconsistency is resolved, but processing complexity increases
Solution Approach 1:
The overlap region processing is extracted and handled separately from the main inference processing flow. By isolating the boundary handling logic into a distinct combining step, the complexity of data synchronization in the neural network intermediate layer is eliminated. The main inference processing remains simple and parallelizable, while boundary consistency is achieved through the separate combining operation.
Data Source
AI summary
An information processing apparatus comprising a division processing unit configured to divide each of input depth data and image data corresponding to the input depth data and having a higher resolution than the input depth data into a plurality of divided regions; an inference processing unit configured to infer depth data by complementing input depth data with image data for each of the divided regions; and a combining processing unit configured to combine depth data having a higher resolution than that of input depth data by combining inferred depth data, wherein the division processing unit performs division such that each divided region has an overlap region partially overlapping with an adjacent divided region.