Planar Target Detection Under Steep Angles Using Warped Image Databases
Solution Overview
Problem
Existing augmented reality systems face challenges in recognizing planar targets from steep angles due to strong out-of-plane rotation issues, which affect the efficiency and effectiveness of descriptor matching, particularly in handling viewpoint and lighting changes.
Innovation Solution
The method involves creating a database by processing both non-warped and warped images of planar targets, detecting keypoints, determining descriptors, and transforming coordinates between different coordinate systems to form a single dataset, allowing for robust recognition of planar targets at various angles, including steep angles, by incorporating warped images to enhance matching capabilities.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If traditional descriptor matching is used for planar target detection, then the system works well for frontal views, but recognition fails at steep angles due to strong out-of-plane rotation
Solution Approach 1:
The patent segments the descriptor matching process into two independent parts: (1) matching the warping parameter to determine the viewpoint angle, and (2) matching the content descriptor based on the determined warping parameter. This segmentation allows the system to handle steep angles by first identifying the rotation angle through warping parameter comparison, then performing content matching adjusted for that specific viewpoint.
Solution Approach 2:
The patent adds a new dimension to the matching process by introducing warping parameter space. Instead of directly comparing content descriptors from different viewpoints, the system first operates in the warping parameter dimension to identify the rotation angle, then uses this information to guide the content matching. This dimensional approach transforms the 3D viewpoint variation problem into a two-step process: angle identification followed by angle-compensated content matching.
2Adaptability or versatility
If the database contains only non-warped images, then the data structure is simple, but the system cannot detect targets at steep angles
Solution Approach 1:
The patent applies preliminary action by pre-computing and storing warping parameters for multiple predetermined rotation angles in the database during offline preparation. These warping parameters (encoding rotation angles like 0°, 30°, 60°, etc.) are calculated in advance and stored alongside or associated with the content descriptors. This preliminary preparation enables the runtime system to quickly identify the closest matching warping parameter without performing complex real-time warping operations.
Solution Approach 2:
The patent changes the database structure by introducing warping parameters as an additional dimension. Each database entry now includes not only the content descriptor but also an associated warping parameter indicating the predetermined rotation angle. This parameter change transforms the database from storing only viewpoint-invariant content descriptors to storing viewpoint-aware descriptors with explicit angle information, enabling efficient steep-angle detection through parameter comparison.
3Reliability
If descriptors are designed to be fully viewpoint invariant, then matching is simpler, but out-of-plane rotation cannot be handled effectively
Solution Approach 1:
The patent makes the descriptor matching process dynamic by determining the warping parameter at runtime based on the captured image characteristics. Instead of using a fixed viewpoint assumption, the system dynamically identifies the rotation angle by comparing the captured image's warping parameters against the predetermined values stored in the database. This dynamic adaptation allows the system to adjust to any steep angle within the predefined range, making the matching process both viewpoint-aware and computationally efficient.
Solution Approach 2:
The patent introduces warping parameters as an intermediary between the captured image and the content descriptor matching. Rather than directly comparing content descriptors from different viewpoints (which fails at steep angles), the system first computes warping parameters from the captured image, uses these as an intermediary to identify the appropriate predetermined rotation angle, and then performs content matching using this intermediate angle information. This intermediary layer bridges the gap between viewpoint variation and content similarity.
Data Source
AI summary
Systems, apparatus and methods to create a database by a device (such as a server) and to use the database by a mobile device for detecting a planar target are presented. The database allows recognition of a planar target by a mobile device from steeper angles with minimum impact on runtime. The database is created from at least one warped view of the planar target. For example, a database may contain keypoints and descriptors from a non-warped view and also from one or more warped views. The database may be pruned by removing keypoints and corresponding descriptors of one image (e.g., a warped image) overlapping with similar or identical keypoints and descriptors of another image (e.g., a non-warped image).