Automated Training Data Collection for Retail Machine Vision

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

The collection of large volumes of training data for machine vision technologies in retail environments is time-consuming, and the deployment of these technologies often requires capturing hundreds or more images of each item type, which can be cumbersome and inefficient.

Innovation Solution

A method and apparatus in a mobile computing device that captures images, tracks the device's pose in a coordinate system, detects items, determines their location, obtains item identifiers, generates training data samples, and stores them, thereby automating the collection of training data while the device is used for other tasks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Quantity of substance

If manual collection of training data is performed to ensure sufficient volume and variety, then the quality and completeness of training data improves, but the time and labor required increases significantly

Engineering Contradiction:

Improvevolume of training dataVSAvoidtime to collect training data

Core Design Contradiction:

Quantity of substanceVSLoss of time

Solution Approach 1:

The system automatically generates training data samples using existing store layout information, item location data, and synthetic image generation techniques. The processor creates labeled training images without human intervention by virtually rendering items in their expected locations based on store plan data, thereby serving the data collection need internally rather than through manual processes

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system pre-generates training data samples before they are actually needed for model training. By using stored store layout information and item location databases, the system can create training datasets in advance, eliminating the need for time-consuming on-site data collection when the training process is initiated

Inventive Principle:

Principle #10Preliminary action

2Reliability

If extensive training data collection is performed to improve model accuracy, then the reliability of item detection improves, but the complexity of the data collection process increases

Engineering Contradiction:

Improveitem detection accuracyVSAvoiddata collection process complexity

Core Design Contradiction:

ReliabilityVSDevice complexity

Solution Approach 1:

The system uses a unified data generation approach that works across multiple store layouts and item types. The same synthetic image generation process, based on store plan data and item location information, can generate training data for any retail environment, eliminating the need for separate collection processes for different scenarios

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system introduces synthetic image generation as an intermediary between raw store data and training model requirements. Instead of directly capturing real images in various conditions, the system uses computational rendering to create realistic training samples, simplifying the data collection pipeline while maintaining data quality

Inventive Principle:

Principle #24Intermediary (Mediator)

3Productivity

If automated training data generation is implemented to reduce manual effort, then the productivity of data collection improves, but the complexity of the system architecture increases

Engineering Contradiction:

Improvetraining data collection efficiencyVSAvoidsystem architecture complexity

Core Design Contradiction:

ProductivityVSDevice complexity

Solution Approach 1:

The system combines multiple data sources (store layout information, item location databases, product catalogs) into a unified training data generation pipeline. By merging these existing systems with the synthetic image generation capability, the solution leverages already-invested infrastructure rather than requiring completely separate automated collection hardware

Inventive Principle:

Principle #5Merging (Combining)

Solution Approach 2:

The system replaces physical data collection mechanisms (cameras, sensors, manual image capture devices) with computational methods. Instead of using mechanical image capture hardware to collect training data in the field, the system uses software-based synthetic image generation driven by store plan data and item location information

Inventive Principle:

Principle #28Mechanics substitution (Replace mechanical system)

Data Source

PatentUS12327371B2Automatic training data sample collection

Publication Date: 2025.06.10 ZEBRA TECHNOLOGIES CORP

AI summary

A method in a mobile computing device includes: controlling a camera to capture an image; tracking a pose of the mobile computing device, corresponding to the image, in a coordinate system; detecting an item in the image; determining a location of the detected item in the coordinate system, based on the tracked pose; obtaining an item identifier corresponding to the detected item, based on the location of the detected item in the coordinate system; generating a training data sample including (i) a payload based on the detected item, and (ii) a label including the obtained item identifier; and storing the training data sample.