Stereo Camera Depth Estimation for Narrow-Baseline Range Extension

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Small robots equipped with stereo cameras face limitations in extending their depth perception range due to mechanical constraints on the baseline length, making it difficult to recognize environments beyond the camera's specified range without increasing the camera's size or cost.

Innovation Solution

A method and apparatus that utilize a depth estimation model to extend the physical recognition range of a stereo camera by generating a depth estimation model based on first and second depth images, allowing estimation of depth values outside the camera's specified range using monocular depth values and a neural network.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If the baseline length of the stereo camera is extended to increase depth perception range, then the depth perception range is improved, but the device size and mechanical complexity increase

Engineering Contradiction:
Improvedepth perception rangeVSAvoidbaseline length
Core Design Contradiction:
Measurement precisionVSLength of moving object

Solution Approach 1:

The patent replaces the mechanical solution of extending the baseline with a computational approach. A depth estimation model (neural network) processes monocular images to predict depth values, substituting the need for physical baseline extension with algorithmic depth estimation. This allows the system to achieve extended depth perception range without increasing the physical dimensions of the stereo camera.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Measurement precision

If a small LiDAR device is used to extend depth perception range, then the depth perception range is improved, but the production cost and installation difficulty increase

Engineering Contradiction:
Improvedepth perception rangeVSAvoidproduction cost
Core Design Contradiction:
Measurement precisionVSEase of manufacture

Solution Approach 1:

The patent creates a virtual copy of LiDAR depth measurement capability through software-based depth estimation. The depth estimation model replicates the depth sensing function of LiDAR by processing standard camera images, providing comparable depth information without requiring expensive LiDAR hardware. This software copy approach maintains depth perception capabilities while dramatically reducing production costs and installation complexity.

Inventive Principle:
Principle #26Copying

3Length of moving object

If the stereo camera baseline is kept narrow for small robot applications, then the device size is reduced, but the depth perception range beyond specification limits is lost

Engineering Contradiction:
Improvebaseline lengthVSAvoiddepth perception range
Core Design Contradiction:
Length of moving objectVSMeasurement precision

Solution Approach 1:

The patent makes the narrow-baseline stereo camera system multi-functional by adding the depth estimation model that processes both stereo images and monocular images. This allows the same hardware configuration to serve dual purposes: maintaining its original stereo vision function while simultaneously achieving extended depth perception range through the integrated depth estimation algorithm, effectively making the system adaptable to both near and far depth requirements.

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

Data Source

PatentUS12524896B2Method and apparatus for estimating depth of image
Publication Date: 2026.01.13 ELECTRONICS & TELECOMM RES INST
  • US12524896B2 patent drawing
  • US12524896B2 patent drawing
  • US12524896B2 patent drawing

AI summary

Disclosed herein is a method for estimating a depth of an image. The method may include acquiring a first depth image based on left and right images collected from a stereo camera, estimating a second depth image based on any one of the left and right images, generating a depth estimation model based on first depth values within a depth range recognizable by the stereo camera in the first depth image and on second depth values of the second depth image matching the first depth values, and inputting third depth values of a range intended to be estimated in the second depth image to the depth estimation model, thereby estimating fourth depth values of the first depth image matching the third depth values.