Specialized U-Disparity Maps for Fast, Accurate Object Recognition

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing technologies for object recognition, particularly in automobiles, face challenges in processing speed and accuracy due to the use of a single U-Disparity map, which can lead to inefficient processing and incorrect object detection.

Innovation Solution

Employing multiple U-Disparity maps tailored to specific image processing objectives to enhance accuracy and speed in object recognition, utilizing a stereo camera system with rectified imaging units to derive parallax values and calculate distances.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If a single U-Disparity map is used for object recognition, then the processing method is simple, but the processing speed decreases and accuracy deteriorates depending on the processing objective

Engineering Contradiction:
Improveprocessing speedVSAvoidcomplexity of U-Disparity map processing
Core Design Contradiction:
ProductivityVSDevice complexity

Solution Approach 1:

The patent segments the U-Disparity map processing by creating multiple specialized maps (first U-Disparity map for object recognition, second U-Disparity map for face detection) instead of using a single general-purpose map. This segmentation allows each map to be optimized for its specific task, improving processing speed and accuracy while maintaining manageable complexity through dedicated processing pipelines.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The system dynamically selects which U-Disparity map to use based on the processing objective. The control unit determines whether to generate and process the first U-Disparity map for object recognition or the second U-Disparity map for face detection, allowing the system to adapt its processing approach in real-time based on requirements.

Inventive Principle:
Principle #15Dynamics

2Measurement precision

If a single U-Disparity map is used for all processing objectives, then the system structure is simple, but the detection accuracy deteriorates for specific tasks

Engineering Contradiction:
Improveobject detection accuracyVSAvoidcomplexity of processing system
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent applies local quality by creating U-Disparity maps with different characteristics suited for specific tasks. The first U-Disparity map is optimized for object recognition with appropriate resolution and processing parameters, while the second U-Disparity map is optimized for face detection with different parameters. Each map has localized quality characteristics matched to its specific detection task.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The processing system is segmented into specialized components: one processing pipeline for object recognition using the first U-Disparity map, and another pipeline for face detection using the second U-Disparity map. This segmentation allows each component to be optimized independently for its specific function, improving overall detection accuracy.

Inventive Principle:
Principle #1Segmentation

3Measurement precision

If multiple U-Disparity maps are generated for different processing objectives, then processing speed and accuracy improve, but the processing time increases

Engineering Contradiction:
Improveobject recognition accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The system uses dynamic control to generate only the necessary U-Disparity map based on the current processing objective. The control unit determines whether to generate the first or second U-Disparity map, avoiding the need to generate both maps simultaneously. This dynamic approach maintains high accuracy while reducing processing time by eliminating unnecessary map generation.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent extracts and separates the processing requirements for different objectives, generating only the specific U-Disparity map needed for the current task. Instead of processing all possible maps regardless of need, the system extracts and processes only the relevant map, reducing unnecessary processing time while maintaining accuracy for the specific task at hand.

Inventive Principle:
Principle #2Taking out (Extraction)

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

Improves the accuracy and processing speed of image processing by utilizing different U-Disparity maps, enabling effective object detection and vehicle control systems.

Implementation Method 1

a parallax value computing unit configured to derive, from a plurality of taken images, a parallax value representing a parallax with respect to the object

Methodology Applied
Scientific EffectParallax: Parallax

Data Source

PatentEP3389009B1Image processing device, object recognition device, apparatus control system, image processing method and program
Publication Date: 2025.08.20 RICOH CO LTD
  • EP3389009B1 patent drawingFigure 1~2(b)
  • EP3389009B1 patent drawingFigure 3
  • EP3389009B1 patent drawingFigure 4(a)~4(b)

AI summary

An image processing device includes a first extracting unit configured to extract a first area representing an object, from a first image indicating a frequency distribution of distance values corresponding to a travelling direction of the image processing device, the frequency distribution associating actual distances in a direction orthogonal to the travelling direction with the distance values; a first processing unit configured to perform first processing to detect a face of the object represented by the first area, using at least the first image; and a second processing unit configured to perform second processing to identify a type of the face of the object represented by the first area, using at least a second image indicating a frequency distribution of the distance values corresponding to the travelling direction of the image processing device, the frequency distribution associating a horizontal direction of a distance image made from the distance values, with the distance values.