Plane Detection in Depth Images via Patch Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for detecting plane models in 3D space, such as RANSAC, are inadequate for environments with multiple planes and are unreliable in the presence of noise, requiring user input and failing to provide optimal solutions.
Innovation Solution
An apparatus and method that divides a depth image into patches, estimates plane models, merges similar patches based on normal vectors, average depth values, and distance thresholds, and refines the models using a consensus set to accurately detect plane models in 3D spaces with multiple planes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If RANSAC method is used to estimate plane models, then the method can work for single-plane environments, but it fails when multiple planes are present in the 3D space
Solution Approach 1:
The depth image is divided into multiple patches, and each patch is processed independently to estimate plane models. This segmentation allows the system to handle multiple planes in different regions of the image without interference, resolving the limitation of RANSAC which assumes a single plane throughout the entire image.
Solution Approach 2:
After estimating plane models for individual patches, the system merges these local plane models into a comprehensive set of plane models for the entire 3D space. This merging process combines the strengths of local plane detection with global plane identification, enabling reliable detection of multiple planes throughout the scene.
2Ease of manufacture
If RANSAC method is used for plane estimation, then the method is simple to implement, but it is unreliable when noise exists in the 3D information
Solution Approach 1:
By dividing the depth image into patches, the system isolates noise within individual patches while maintaining the overall structure. Each patch can be processed independently with plane estimation algorithms that are less sensitive to noise, and the results are combined to form reliable plane models for the entire scene.
Solution Approach 2:
The system uses a feedback mechanism where plane models estimated from patches are validated and refined through the merging process. Patches with inconsistent or noise-corrupted plane models are identified and reprocessed, gradually improving the reliability of the final plane detection results.
3Adaptability or versatility
If variation methods such as sequential RANSAC or multi-RANSAC are used, then the method can handle multiple planes, but it requires user to pre-designate the number of planes and is unreliable with noise
Solution Approach 1:
The system automatically determines the number and location of planes through the patch-based estimation and merging process. Instead of requiring users to pre-specify the number of planes, the algorithm self-determines this parameter by analyzing the structural characteristics of the depth image and the consistency of plane models across patches.
Solution Approach 2:
The automatic plane detection is achieved through segmentation into patches, where each patch independently estimates local plane models. The merging of these patch-level models automatically reveals the global plane structure, eliminating the need for user input while maintaining reliability in noisy conditions.
Data Source
AI summary
A plane detection apparatus for detecting at least one plane model from an input depth image. The plane detection apparatus may include an image divider to divide the input depth image into a plurality of patches, a plane model estimator to calculate one or more plane models with respect to the plurality of patches including a first patch and a second patch, and a patch merger to iteratively merge patches having a plane model a similarity greater than or equal to a first threshold by comparing plane models of the plurality of patches. When a patch having the plane model similarity greater than or equal to the first threshold is absent, the plane detection apparatus may determine at least one final plane model with respect to the input depth image using previously merged patches.