Plane Detection via Depth Map Fusion for AR Boundary Alignment
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing plane detection methods fail to detect planes in texture-less regions and cannot align detected planes with the boundaries of real objects, leading to inaccurate results in augmented reality applications.
Innovation Solution
A method and device using a deep neural network to acquire input images, estimate depth maps, and perform region segmentation, which includes a feature extractor, depth estimation branch, and normal estimation branch to optimize depth information and align plane boundaries with real objects through feature map fusion and attention mechanisms.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional plane detection methods are used, then the detection process is simple, but planes in texture-less regions cannot be detected and boundaries are not aligned with real objects
Solution Approach 1:
The patent transitions from 2D image-based plane detection to 3D depth map-based detection. By estimating depth information and working in three-dimensional space, the system can detect planes in texture-less regions where traditional 2D methods fail, as depth provides additional geometric constraints for plane identification and boundary alignment.
Solution Approach 2:
The patent introduces depth maps as an intermediary representation between the input image and plane detection results. The depth map serves as a mediator that captures geometric information, enabling more accurate plane detection and boundary alignment while maintaining a structured detection pipeline that combines traditional processing with deep learning components.
2Measurement precision
If deep neural networks with multiple branches are used to optimize depth estimation, then plane detection accuracy improves, but computational complexity increases
Solution Approach 1:
The patent divides the deep neural network into multiple specialized branches: a depth estimation branch and a normal estimation branch. Each branch focuses on estimating specific geometric properties, allowing the system to achieve high accuracy in depth and normal estimation while maintaining modular computation that can be optimized and executed efficiently.
Solution Approach 2:
The deep neural network is designed to perform multiple functions simultaneously: it estimates depth maps, normal maps, and ultimately detects planes and their boundaries. This multi-functional architecture allows a single model to handle various geometric estimation tasks, improving overall system efficiency and reducing the need for separate specialized models.
Data Source
AI summary
A method and a device for performing plane detection are provided. The method includes acquiring an input image, extracting features of the input image using a deep neural network and estimating a depth map of the input image based on the extracted features, performing region segmentation using the depth map to detect plane regions in the input image.