Object Recognition System Using Dynamic Model Selection

Resolve Bottlenecks,
Find Innovative SolutionsGenerate Solutions

Solution Overview

Problem

Existing object recognition systems face challenges in accurately identifying unknown or temporarily installed objects, especially when three-dimensional map data is outdated, leading to difficulties in recognizing stationary objects and objects with uncertain shapes or observation directions.

Innovation Solution

An object recognition system that includes a recognition device and a server, where the device transmits type-indeterminable images to the server for further analysis using a learning model generated based on training data with teacher labels, allowing for improved object type determination.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Adaptability or versatility

If the three-dimensional map data is not updated, then the recognition device can operate with existing data, but it cannot detect newly installed stationary objects or temporarily installed objects

Engineering Contradiction:

Improverecognition of new objectsVSAvoidtime for map data updates

Core Design Contradiction:

Adaptability or versatilityVSLoss of time

Solution Approach 1:

The system enables automatic recognition and classification of objects without requiring manual map updates. The recognition device autonomously identifies stationary objects, determines their types using multiple learning models, and adapts to new objects in the environment, making the system self-updating rather than relying on external map data maintenance

Inventive Principle:

Principle #25Self-service

Solution Approach 2:

The system dynamically selects and switches between different learning models based on the characteristics of the detected object. When an object is detected, the system determines its type using appropriate learning models and can adapt its recognition approach in real-time, allowing it to handle newly installed objects without pre-existing map data

Inventive Principle:

Principle #15Dynamics

2Measurement precision

If the recognition device uses a fixed learning model, then the device complexity is reduced, but the recognition accuracy for diverse object types and observation directions deteriorates

Engineering Contradiction:

Improveobject type recognition accuracyVSAvoidmultiple learning models

Core Design Contradiction:

Measurement precisionVSDevice complexity

Solution Approach 1:

The recognition device incorporates multiple learning models that can handle different object types, shapes, and observation directions. Each learning model is specialized for certain object characteristics, and the system selects the appropriate model based on the detected object, providing universal recognition capability across diverse scenarios without requiring a single complex model

Inventive Principle:

Principle #6Universality (Multi-functionality)

Solution Approach 2:

The system dynamically selects which learning model to use based on the characteristics of the detected object. This dynamic model selection allows the system to adapt to different object types and observation conditions, improving recognition accuracy while managing complexity through intelligent model choice rather than using all models simultaneously

Inventive Principle:

Principle #15Dynamics

3Measurement precision

If the system transmits all type-indeterminable images to the server, then the recognition accuracy improves through learning model updates, but the communication load and processing time increase

Engineering Contradiction:

Improverecognition accuracyVSAvoidprocessing efficiency

Core Design Contradiction:

Measurement precisionVSProductivity

Solution Approach 1:

The system transmits only a subset of type-indeterminable images to the server for learning model updates, rather than all possible images. This partial action approach focuses on transmitting images that are most valuable for improving recognition accuracy, thereby reducing communication load and processing time while still achieving the goal of improving recognition performance

Inventive Principle:

Principle #16Partial or excessive action

Data Source

PatentUS11783602B2Object recognition system, recognition device, object recognition method, and object recognition program

Publication Date: 2023.10.10 NEC CORP

AI summary

An object recognition system 80 includes: a recognition device 30 that recognizes an object in an image; and a server 40 that generates a learning model. The recognition device 30 includes: a first object recognition unit 310 that determines a type of the object in the image using the learning model; and an image transmission unit 320 that transmits a type-indeterminable image, which is an image in which the type has not been determined, to the server 40 when an object included in the type-indeterminable image is an object detected as a three-dimensional object. The server 40 includes: a learning device 410 that generates the learning model based on training data in which a teacher label is assigned to the type-indeterminable image; and a learning model transmission unit 420 that transmits the generated learning model to the recognition device 30. The first object recognition unit 310 determines the type of the object in the image using the transmitted learning model.