Feature Detection Using Statistical Template Detectors

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods for locating features in images, such as Active Shape Models and Active Appearance Models, face challenges in robust detection and tracking across sequences without significant increases in computational complexity.

Innovation Solution

The method generates a set of template detectors using a statistical model, iteratively improving feature locations in a target image by correlating image patches with the model, and using non-linear optimization to ensure accurate feature point placement, while maintaining computational efficiency.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If conventional methods (ASM/AAM) are used to locate features, then feature detection can be performed, but robust detection and tracking across image sequences cannot be achieved without significant increase in computational complexity

Engineering Contradiction:
Improverobust detection and trackingVSAvoidcomputational complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The method segments the feature detection process into multiple stages: (1) extracting local image patches around initial feature estimates, (2) classifying patches into templates based on appearance models, (3) iteratively refining feature locations by matching templates with image regions. This segmentation allows robust tracking across sequences while managing computational complexity through staged processing.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The method dynamically adapts the feature detection process based on tracking progress. Initial feature points are refined iteratively by comparing template detectors with image patches, and the process adjusts to image sequences by maintaining updated feature locations across frames. This dynamic approach enables robust tracking without requiring excessive computational complexity for each individual frame.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If iterative refinement with template detectors is applied, then feature localization accuracy improves, but computational time increases

Engineering Contradiction:
Improvefeature localization accuracyVSAvoidcomputational time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The method performs preliminary actions by extracting and classifying image patches into templates before the main refinement process. Templates are pre-computed and stored, allowing the iterative refinement to directly match pre-processed templates with image regions rather than processing raw images throughout, thus reducing computational time while maintaining accuracy.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The method implements feedback by iteratively comparing template detectors with image patches and using the results to refine feature locations. The feedback loop adjusts feature estimates based on template matching results, progressively improving accuracy. This feedback mechanism allows the system to achieve high precision while limiting computational time by stopping iterations when convergence is achieved.

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP2005367B1Method of locating features of an object
Publication Date: 2014.07.09 TOYOTA MOTOR EUROPE NV SA
  • EP2005367B1 patent drawingFigure 1
  • EP2005367B1 patent drawingFigure 2
  • EP2005367B1 patent drawingFigure 3

AI summary

A method of locating features of an object, of a class of objects, within a target image. The method comprises initialising a set of feature points within the target image, each feature point corresponding to a predetermined feature for objects of the class of objects; deriving a set of template detectors, from the set of feature points, using a statistical model of the class of objects, each template detector comprising an area of image located about the location of a feature point for an object of the class of objects; comparing the set of template detectors with the target image; and updating the set of feature points within the target image in response to the result of the comparison.