Stereo Object Detection Using Reliability-Guided Disparity Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object detection devices face challenges in maintaining disparity calculation accuracy and preventing performance degradation due to vertical-direction image deviation, leading to potential undetection and increased processing costs.

Innovation Solution

The device employs a reliability calculation execution determination unit to assess the necessity of disparity calculation based on distance thresholds, movement speed, and curvature, and an object detection unit that classifies disparities by reliability, ensuring accurate object detection while minimizing unnecessary processing.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If straight-line detection and warning processing are executed for all viewpoint images, then disparity calculation accuracy is improved, but processing cost increases and objects may not be detected

Engineering Contradiction:
Improvedisparity calculation accuracyVSAvoidprocessing efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent changes the parameter of processing selection by introducing a reliability flag that determines whether disparity calculation should be performed for each viewpoint image pair. This flag is set based on the angle θ between straight lines and horizontal lines, creating a conditional parameter change that optimizes both accuracy and efficiency.

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent applies local quality by treating different viewpoint image pairs differently based on their specific geometric characteristics. Instead of uniform processing, the system selectively applies disparity calculation only to image pairs where the angle θ exceeds the threshold, creating local quality differentiation in processing intensity.

Inventive Principle:
Principle #3Local quality

2Reliability

If disparity calculation is performed without selective filtering, then all objects are detected, but processing cost grows and performance degrades

Engineering Contradiction:
Improveobject detection accuracyVSAvoidprocessing complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent performs preliminary action by calculating the angle θ and setting the reliability flag before executing disparity calculation. This preliminary assessment prevents unnecessary processing of viewpoint image pairs with small angles, reducing overall processing complexity while maintaining detection reliability.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent extracts and isolates the problematic case by identifying viewpoint image pairs with small angles θ and excluding them from disparity calculation through the reliability flag mechanism. This extraction of unnecessary processing cases reduces device complexity and processing load.

Inventive Principle:
Principle #2Taking out (Extraction)

Data Source

PatentEP3825648B1Object detection device
Publication Date: 2026.04.01 ASTEMO LTD
  • EP3825648B1 patent drawingFigure 1
  • EP3825648B1 patent drawingFigure 2~3
  • EP3825648B1 patent drawingFigure 4

AI summary

The present invention ensures disparity calculation accuracy even if there is vertical-direction image deviation and suppresses performance degradation and processing cost growth. The present invention comprises: an imaging unit 310 for obtaining a plurality of images; a disparity image generation unit for generating a disparity image from images acquired from the imaging unit 310; a reliability calculation unit 342 for using inter-matching-window edge orientations to calculate the reliabilities of the disparities in the disparity image; a reliability calculation execution determination unit 341 for determining whether to execute reliability calculation; and an object detection unit 343 for, if the reliability calculation execution determination unit 341 determines that reliability calculation is to be executed, using the disparity image and reliabilities to detect an object.