Objectness Estimation Using Depth Edge Density and Uniformity
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing objectness estimation methods struggle to accurately differentiate between regions containing single objects and those with multiple objects or patterns, leading to increased objectness scores in inadequately captured regions, especially when relying on visible light images or depth images without learning data.
Innovation Solution
An objectness estimation apparatus and method that calculates objectness based on edge density and uniformity on the periphery and inside of candidate regions in depth images, using edge detection and integral image generation to accurately assess the presence of a single object, without relying on learning data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If objectness is calculated based on visible light image boundaries (edge detection or super-pixel division), then objectness estimation can be performed, but boundaries formed from object patterns are incorrectly identified as object boundaries, leading to inaccurate objectness scores in regions with inadequately captured objects
Solution Approach 1:
The patent introduces depth information as an intermediary to distinguish true object boundaries from pattern boundaries. Depth edges detected from depth images serve as a mediator to validate visible light image boundaries, allowing the system to differentiate between boundaries caused by actual object boundaries versus those caused by surface patterns or textures.
Solution Approach 2:
The patent combines visible light image information with depth image information to create a composite boundary detection system. By integrating edge detection results from both visible light and depth images, the system achieves more accurate object boundary identification that overcomes the limitations of using either image type alone.
2Reliability
If high objectness is calculated in regions including inaccurately captured objects, then more candidate regions are output to cover all objects, but this increases calculation cost and causes erroneous recognition
Solution Approach 1:
Depth edge information serves as an intermediary validation mechanism to filter out false positive candidate regions. By requiring both visible light image boundaries and depth image boundaries to coincide, the system reliably identifies true object boundaries without needing to process excessive candidate regions, thus maintaining detection accuracy while improving efficiency.
Solution Approach 2:
The patent changes the parameters used for objectness calculation by incorporating depth edge density and uniformity metrics. This parameter change allows the system to more accurately assess whether a candidate region contains a complete object, reducing the number of false positives and improving both reliability and productivity.
3Measurement precision
If objectness estimation uses learning data, then accuracy can be improved for known objects, but the system cannot handle unknown objects or obstacles not included in the learning data
Solution Approach 1:
The patent employs a self-service approach by using geometric and depth-based features that are inherently present in the image data itself, rather than relying on external learning data. The system uses edge detection, depth edge density calculation, and boundary coincidence checks that work for any object type, making the system both accurate and universally adaptable without requiring training on specific object classes.
Data Source
AI summary
Objectness indicating a degree of accuracy of a single object is accurately estimated. An edge detection unit 30 detects an edge for a depth image, an edge density/uniformity calculation unit 40 calculates an edge density on the periphery of a candidate region, an edge density inside the candidate region, and edge uniformity on the periphery of the candidate region. An objectness calculation unit 42 calculates the objectness of the candidate region based on the edge density on the periphery of the candidate region, the edge density inside the candidate region, and the edge uniformity on the periphery of the candidate region.