Robot Object Detection Model Segmentation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing robot systems face challenges in reliably detecting objects due to changes in shadow, pose, texture, and surrounding objects, which affect pattern matching processing, leading to unsuccessful object detection.
Innovation Solution
A robot system that includes a processing apparatus capable of selecting and generating a model based on two images captured under different conditions, using feature extraction, model integration, and matching to exclude areas with different features, allowing for reliable object detection by focusing on unchanged features and position/pose relationships.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If pattern matching is performed using a single registered model, then the processing is simple and fast, but detection reliability deteriorates due to changes in shadow, pose, texture, and surrounding objects
Solution Approach 1:
The patent segments the object model into multiple regions (first region and second region) with different feature extraction strategies. The first region uses features that are invariant to imaging conditions, while the second region uses features that capture condition-dependent variations. This segmentation allows the system to maintain reliability across varying conditions without requiring completely separate models for each condition.
Solution Approach 2:
The patent changes the parameters used for feature extraction based on the region being analyzed. For the first region, parameters are selected that remain stable across different imaging conditions, while for the second region, parameters capture the variations. This parameter adaptation enables the single model to handle multiple imaging conditions effectively.
2Reliability
If multiple models are registered to account for different imaging conditions, then detection reliability improves, but the complexity of model management and processing increases
Solution Approach 1:
The patent divides the pattern matching process into two stages corresponding to the two regions. The first stage processes the first region using condition-invariant features, and the second stage processes the second region using condition-dependent features. This segmentation allows parallel processing of different feature types, reducing overall processing time compared to sequential model evaluation.
Solution Approach 2:
The patent applies partial pattern matching by focusing on the first region with high-priority condition-invariant features for initial matching, then selectively applying the second region's condition-dependent features only when needed. This partial action approach reduces processing time by avoiding full evaluation of all features in all conditions.
3Adaptability or versatility
If features from multiple images are integrated, then the model becomes more robust to imaging condition changes, but the complexity of feature extraction and integration increases
Solution Approach 1:
The patent segments feature extraction into two distinct pipelines: one for the first region extracting condition-invariant features, and another for the second region extracting condition-dependent features. This segmentation simplifies the integration process by clearly defining what features to combine and how, reducing the overall complexity compared to uniform feature extraction across the entire object.
Solution Approach 2:
The patent applies different feature extraction qualities to different regions. The first region uses high-stability, condition-invariant features, while the second region uses features that adapt to local imaging conditions. This local differentiation optimizes adaptability without requiring complex uniform processing across the entire object.
Data Source
AI summary
A robot system includes a processing apparatus that detects one work from a plurality of works, and a robot that operates the detected one work. The processing apparatus includes a display unit that displays image data containing an image of the plurality of works captured by an imaging apparatus, a selection unit that selects a first image and a second image from the image data, and a processing unit that generates a model based on the first image and the second image and detects the one work using the generated model.