Object Counting via Learned Estimation Parameters
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing techniques struggle to accurately count objects, especially when they are partly or completely hidden, such as crops obscured by leaves.
Innovation Solution
An information processing apparatus that includes a feature acquisition unit to detect target objects from images, a selection unit to choose estimation parameters based on image capturing targets, and an estimation unit to estimate the number of target objects using the selected parameters and features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If image processing techniques are used to detect objects, then detection accuracy is improved for visible objects, but detection fails completely when objects are hidden
Solution Approach 1:
The patent introduces an estimation mechanism that acts as an intermediary between detection results and final object count. When detection fails or returns unreliable results (especially for hidden objects), the estimation unit uses learned parameters and features to infer the actual number of objects, bridging the gap between what can be detected and what actually exists
Solution Approach 2:
The patent changes the approach from direct detection to parameter-based estimation. By learning parameters that correlate with object presence (such as leaf density, image features, environmental conditions) and using these parameters to estimate object counts, the system can infer hidden objects without directly detecting them
2Loss of information
If detection processing is applied to all regions, then complete object information is obtained, but processing time and computational resources increase significantly
Solution Approach 1:
The patent applies detection processing selectively rather than uniformly to all regions. By using estimation parameters and features to identify regions where detection is likely to succeed, the system focuses computational resources on promising areas while using estimation for regions where objects are likely hidden, achieving reasonable accuracy with reduced processing
Solution Approach 2:
The patent performs preliminary estimation using learned parameters before committing to full detection processing. This preliminary action identifies regions where detection is worthwhile and regions where estimation should be used, avoiding unnecessary computational expenditure on regions where objects are clearly hidden
Data Source
AI summary
An estimation parameter for estimating the actual number of target objects in a designated region of a field is learned using a feature amount acquired from a captured image of a set region of the field and the actual number of target objects in the set region as learning data.