Object Detection Using Reference Point Feature Tables
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing object detection methods using Hough transform require excessive processing due to a large number of votes in Hough spaces, leading to increased processing demands and decreased accuracy.
Innovation Solution
An object detection method that sets external and internal reference points, calculates feature quantities on a local surface, and uses a table to associate these quantities with positions, excluding inappropriate position vectors to reduce processing and improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If Hough transform is used to detect object position and attitude, then detection capability is achieved, but the amount of processing excessively increases due to enormous number of votes in Hough spaces
Solution Approach 1:
The patent segments the continuous Hough space into discrete regions by introducing a region determination step. Instead of processing all possible votes in the entire Hough space, the method divides the space into multiple regions and processes only relevant regions, thereby reducing the computational burden while maintaining detection accuracy.
Solution Approach 2:
The patent extracts only the necessary votes from Hough spaces that fall within determined regions, rather than processing all votes. By extracting and processing only the relevant subset of votes, the method significantly reduces the amount of processing required while still achieving accurate object position and attitude detection.
2Reliability
If all position vectors from Hough spaces are used for estimation, then comprehensive information is available, but detection accuracy decreases due to inclusion of inappropriate position vectors
Solution Approach 1:
The patent applies local quality by evaluating the validity of position vectors based on their local geometric relationships with reference points. Instead of treating all position vectors uniformly, the method assesses each vector's validity based on its specific spatial relationship with object reference points, thereby filtering out inappropriate vectors and improving detection accuracy.
Solution Approach 2:
The patent implements a feedback mechanism where the validity of position vectors is continuously evaluated based on geometric relationships with reference points. The system provides feedback by determining whether each position vector is appropriate or inappropriate, and adjusts the estimation process accordingly, thereby improving reliability and accuracy.
Data Source
Figure 1A~1B
Figure 2
Figure 3
AI summary
In an object detecting method according to an embodiment, external reference points (O1, O2, O3, O5, O6, O7) are set in external space of a model (M) of an object (W) and an internal reference point (O4, O8, O9) is set in internal space of the model (M). A table (81) is stored in which feature quantities on a local surface of the model (M) are associated with positions of the external reference points (O1, O2, O3, O5, O6, O7) and the internal reference point (O4, O8, O9). The feature quantity on the local surface of the model (M) is calculated, and the position of the reference point whose feature quantity is identical to the calculated feature quantity is acquired from the table and is converted into a position in a real space. When the converted position is outside the object (W), the position is excluded from information for estimation and the position and the attitude of the object (W) are estimated.