Object Tracking With Neural Template Feature Extraction

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object tracking methods in moving images rely on manually designed template features, which are not effectively evaluated for tracking suitability and require numerous patterns for arbitrary objects, leading to inefficiencies and potential tracking failures.

Innovation Solution

An information processing apparatus that uses a feature extraction neural network to automatically generate and identify template features for object tracking, incorporating a determination unit to determine tracking positions, generation units for reference and search images, and an identification unit for correlation operations to enhance tracking accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Ease of manufacture

If manual template feature design is used, then tracking can be performed with simple methods, but tracking reliability deteriorates because manually designed features are not effectively evaluated for tracking suitability

Engineering Contradiction:
Improveease of template feature designVSAvoidtracking reliability
Core Design Contradiction:
Ease of manufactureVSReliability

Solution Approach 1:

The system uses the object itself to generate its own template features through automatic feature extraction, eliminating the need for manual feature design. The feature extraction unit automatically extracts features from the reference image based on the object's actual appearance, allowing the system to adapt to any object without manual intervention.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent changes the parameter of template feature generation from manual design to automatic extraction using neural networks. By using a feature extraction unit that processes images through learned parameters, the system can generate optimal features for any object without requiring hand-crafted templates.

Inventive Principle:
Principle #35Parameter changes

2Adaptability or versatility

If multiple template patterns are prepared for arbitrary objects, then tracking adaptability improves, but device complexity increases due to the enormous number of template patterns required

Engineering Contradiction:
Improvetracking adaptability to arbitrary objectsVSAvoidcomplexity of template pattern collection
Core Design Contradiction:
Adaptability or versatilityVSDevice complexity

Solution Approach 1:

The feature extraction unit serves as a universal system that can extract appropriate template features for any arbitrary object by inputting a reference image. Instead of requiring separate template patterns for different objects, the same automated extraction system adapts to generate suitable features for any object type, eliminating the need for extensive template libraries.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Speed

If manually designed template features are used, then processing speed is maintained at simple levels, but tracking precision deteriorates because suitable feature combinations are not examined

Engineering Contradiction:
Improveprocessing speedVSAvoidtracking precision
Core Design Contradiction:
SpeedVSMeasurement precision

Solution Approach 1:

The system performs preliminary feature extraction and generation before actual tracking occurs. By pre-extracting features from the reference image and storing them as template features, the system prepares accurate feature representations in advance, enabling precise tracking without requiring complex real-time processing.

Inventive Principle:
Principle #10Preliminary action

Data Source

PatentUS12444059B2Information processing apparatus, learning apparatus, and tracking method
Publication Date: 2025.10.14 CANON KK
  • US12444059B2 patent drawing
  • US12444059B2 patent drawing
  • US12444059B2 patent drawing

AI summary

An information processing apparatus obtains a reference image including a target object image to be tracked, determines a plurality of positions in the reference image, and generates a plurality of first features corresponding to the plurality of positions by inputting the reference image to a feature extraction neural network. The apparatus further obtains a search image that is a target of the tracking processing of the target object image, and generates a second feature by inputting the search image to the feature extraction neural network. Then, the apparatus identifies a position of the target object image included in the search image based on a result of a correlation operation between each of the plurality of first features and the second feature.