Vector Neural Network Object Detection for Transparent Objects

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing object detection techniques, such as template matching, struggle to detect objects when templates are not available or when objects are transparent, as they rely on pre-defined templates for recognition.

Innovation Solution

The use of a vector neural network type machine learning model with multiple vector neuron layers to detect objects from input images by generating a similarity image and a discriminant image, where the similarity indicates the degree of similarity to known features and the discriminant image assigns an unknown label to pixels with low similarity.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If template matching is used for object detection, then high-speed object recognition can be achieved, but it becomes impossible to detect objects when templates are not available or when objects are transparent

Engineering Contradiction:

Improveobject recognition speedVSAvoiddetection capability for unknown or transparent objects

Core Design Contradiction:

ProductivityVSAdaptability or versatility

Solution Approach 1:

The patent replaces the mechanical template matching system with a neural network-based learning system. Instead of mechanically comparing input images against pre-defined templates, the system uses a neural network that has learned object features from training data, enabling it to detect both known and unknown objects including transparent ones without requiring manual template preparation.

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The patent changes the fundamental parameters of object detection by transitioning from template-based comparison to neural network feature extraction. The neural network learns optimal feature representations during training, automatically adapting to different object types and conditions without requiring manual template creation, thus solving the limitation of detecting only pre-defined objects.

Inventive Principle:

Principle #35Parameter changes

2Device complexity

If traditional template matching methods are used, then simple and fast detection can be performed, but the method fails when objects do not match predefined templates

Engineering Contradiction:

Improvedetection method simplicityVSAvoiddetection accuracy for varied objects

Core Design Contradiction:

Device complexityVSReliability

Solution Approach 1:

The patent performs preliminary learning action during a training phase where the neural network learns object features from labeled training images before actual detection. This preliminary learning enables the system to automatically adapt to various object types and conditions, ensuring reliable detection without requiring manual template preparation for each object class.

Inventive Principle:

Principle #10Preliminary action

Solution Approach 2:

The neural network performs self-learning by automatically extracting and learning object features from training data without human intervention. The system self-adjusts its parameters and feature representations during training, enabling it to reliably detect varied objects without requiring manual template creation or adjustment for each object type.

Inventive Principle:

Principle #25Self-service

Data Source

PatentUS12293578B2Object detection method, object detection apparatus, and non-transitory computer-readable storage medium storing computer program

Publication Date: 2025.05.06 SEIKO EPSON CORP

AI summary

An object detection method includes inputting an input image to a learned machine learning model and generating a similarity image from an output of at least one specific layer, and generating a discriminant image to which at least an unknown label is assigned, by comparing a similarity of each pixel in the similarity image to a predetermined threshold value.