Stereo Vision Depth Estimation Using Adaptive Filter Shapes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Traditional stereoscopic vision methods face challenges in depth estimation due to the violation of the fronto-parallel assumption and low texture information, leading to poor correlations, especially when dealing with non-frontal parallel surfaces and complex scenes.

Innovation Solution

The method employs a multiplicative combination of differently shaped and sized matching filters to improve depth estimation by selecting the best filter shape for each image position, using a statistical approach to combine correlation values from various filters, thereby overcoming the limitations of traditional block-matching methods.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If traditional block-matching methods use a single filter shape for depth estimation, then the device complexity is low, but the depth estimation accuracy deteriorates when dealing with non-frontal parallel surfaces and complex scenes

Engineering Contradiction:
Improvedepth estimation accuracyVSAvoidfilter combination complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the depth estimation process by applying multiple filtering operations with different shapes (e.g., horizontal, vertical, oblique filters) to different directional components of the stereo images. Each filter processes specific directional features independently, and the results are combined to achieve comprehensive depth estimation. This segmentation allows the system to handle complex scenes with non-frontal surfaces by capturing directional variations separately and integrating them.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements local quality by adapting the filter shape and orientation to match the local surface characteristics at each image position. By analyzing the local image structure and selecting filters that align with the predominant surface orientation in each region, the system optimizes depth estimation accuracy for locally varying geometries. This local adaptation enables robust performance on surfaces with varying orientations without requiring a single complex global filter.

Inventive Principle:
Principle #3Local quality

2Measurement precision

If traditional stereoscopic methods assume fronto-parallel surfaces, then the processing is simple and fast, but the measurement precision deteriorates when the assumption is violated

Engineering Contradiction:
Improvedepth estimation accuracy on non-frontal surfacesVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces dynamics by making the filter orientation adaptive rather than fixed. The system dynamically selects and applies filters with different orientations based on the local image content and estimated surface normals. This dynamic adaptation allows the processing to automatically adjust to non-frontal surfaces, capturing the varying orientations throughout the scene. The dynamic filter selection mechanism enables accurate depth estimation on complex geometries while maintaining a modular processing framework.

Inventive Principle:
Principle #15Dynamics

3Measurement precision

If multiple differently shaped filters are used for each image position, then the depth estimation accuracy improves, but the computational time and energy consumption increase

Engineering Contradiction:
Improvedepth estimation accuracyVSAvoidprocessing time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent applies partial action by selectively using different filter shapes based on the local image characteristics and confidence metrics. Rather than uniformly applying all possible filter combinations across the entire image, the system adapts the filtering strategy to each region's needs. In areas with clear surface orientations, fewer targeted filters are sufficient, while regions with ambiguous or complex structures receive more comprehensive filtering. This selective application reduces overall computational load while maintaining accuracy where it matters most.

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP2960858B1Sensor system for determining distance information based on stereoscopic images
Publication Date: 2018.08.08 HONDA RES INST EUROPE
  • EP2960858B1 patent drawingFigure 1~2
  • EP2960858B1 patent drawingFigure 3~4
  • EP2960858B1 patent drawingFigure 5~6

AI summary

The invention provides a distance measurement method determining the distance of a sensor system to a physical object, comprising the steps of obtaining, from the sensor system, at least a pair of stereoscopic images including the physical object, applying to each element of at least a portion of a first image of the pair of stereoscopic images and to each element of at least a portion of a second image of the pair of stereoscopic images at least two differently shaped and/or sized filters, respectively, determining correlation values for each filter applied to the first and second image, determining combined correlation values for the applied filters by combining the determined correlation values for each applied filter, evaluating the combined correlation values for different disparities for an extremum value of the combined correlation values, calculating a distance value of the sensor system to the physical object based on a disparity value at which the extremum occurs, and outputting the distance value.