Carried Item Region Extraction Using Stable String Features

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing methods struggle to accurately extract carried items from image information due to variations in surface texture, color, shape, and posture, as well as the need for extensive data preparation and constant positional relationships, which are not always feasible.

Innovation Solution

A carried item region extraction device that extracts a string region with a stable shape from image information and uses this region to identify and extract the carried item, eliminating the need for extensive data preparation and accounting for shape changes and posture variations.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If template matching or SIFT/HOG methods are used to extract carried items, then extraction accuracy can be improved for specific object types, but extensive data preparation and multiple templates are required for different appearances

Engineering Contradiction:

Improveextraction accuracyVSAvoiddata preparation complexity

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The patent extracts the string region (shoulder strap, handle, or carrying portion) as a separate feature from the main carried item body. This extracted string region serves as a stable reference point that simplifies the overall extraction process, allowing the system to locate carried items without requiring extensive templates for different item appearances

Inventive Principle:

Principle #2Taking out (Extraction)

Solution Approach 2:

The string region detection method serves multiple purposes: it identifies the presence of a carried item, determines its position, and provides a reference for further analysis. This single feature extraction approach works across different carried item types (bags, backpacks, handbags) without requiring type-specific templates

Inventive Principle:

Principle #6Universality (Multi-functionality)

2Adaptability or versatility

If traditional methods detect carried items based on overall shape and texture, then comprehensive item recognition is achieved, but accuracy decreases due to shape changes from posture variations

Engineering Contradiction:

Improveitem type coverageVSAvoidextraction accuracy

Core Design Contradiction:

Adaptability or versatilityVSMeasurement precision

Solution Approach 1:

The patent divides the carried item into two segments: the string region (shoulder strap or handle) and the main body. By focusing detection on the string region which maintains a relatively stable appearance, the system achieves accurate detection while the main body can vary in shape due to different item types and carrying postures

Inventive Principle:

Principle #1Segmentation

Data Source

PatentUS8977005B2Carried item region extraction device, carried item region extraction method, and computer readable medium

Publication Date: 2015.03.10 NEC CORP

AI summary

Provided is a carried item region extraction device for accurately extracting a carried item region from an image.This carried item region extraction device has: a string region processing unit for extracting a string region including a string of a carried item from image information; and a carried item region processing unit for extracting a carried item region including a carried item from the image information on the basis of the string region.