Depth Map Correction Using Object Size Thresholds
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing depth map estimation techniques, such as the stereo method, depth from defocus (DFD), and depth from focus (DFF), suffer from errors, especially in areas with little texture, leading to inaccurate distance calculations and increased measurement errors, and previous correction methods do not adequately consider distance estimation accuracy or account for object size and texture.
Innovation Solution
An image processing apparatus and method that detects objects in an image, determines their size relative to a threshold, and corrects distance values in areas with low estimation accuracy by using a representative value, such as a mean, median, or mode, to improve depth map accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If clustering processing is performed to correct depth map without considering distance estimation accuracy, then correction processing can be applied uniformly, but correction is performed even for areas with accurate distance estimation, increasing deviation from correct distance
Solution Approach 1:
The patent applies different correction strategies to different regions of the depth map based on local characteristics. Areas with low texture (poor estimation accuracy) receive clustering-based correction, while areas with high texture (good estimation accuracy) are left uncorrected. This local differentiation resolves the contradiction by ensuring correction is only applied where needed, maintaining both measurement precision and reliability.
2Measurement precision
If clustering is used to divide local areas for correction, then correction can be applied to regions, but correction to make the distance of each object the same cannot be performed
Solution Approach 1:
The patent segments the depth map into multiple candidate regions using clustering, then performs object detection within each region. This allows the system to identify individual objects and apply object-specific corrections, making the correction process adaptable to different objects while maintaining regional processing efficiency.
Solution Approach 2:
The patent dynamically adjusts the correction approach based on detected object characteristics. For each detected object, the system determines whether to apply uniform regional correction or object-specific correction based on the object's properties, enabling versatile adaptation to different object types and sizes.
3Reliability
If distance correction is applied to all areas, then comprehensive correction is achieved, but unnecessary corrections increase deviation from correct distances in accurately estimated areas
Solution Approach 1:
The patent changes the parameter of correction application from uniform (all areas) to conditional (only low-texture areas). By using texture analysis to identify areas with poor distance estimation accuracy, the system selectively applies correction only where needed, maintaining overall correction coverage while preserving accuracy in well-estimated areas.
4Device complexity
If object size is not considered in correction processing, then processing is simplified, but correction cannot be adapted to different object sizes and textures
Solution Approach 1:
The patent performs preliminary object detection and size classification before applying correction processing. By detecting objects and categorizing them by size and texture characteristics in advance, the system prepares object-specific correction parameters, enabling adaptable correction without significantly increasing processing complexity during the main correction phase.
Data Source
AI summary
An image processing apparatus for correcting, on the basis of an image and a depth map corresponding to the image, the depth map, includes: a detection unit that detects an object included in the image; a determination unit that determines whether a size of the object detected by the detection unit is a threshold or less; and a correction unit that corrects a distance in a target area which corresponds to an area of the object in the depth map, when the size of the detected object is the threshold or less.