Monocular Depth Estimation Using Intermediate Depth Map Refinement

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current unsupervised depth estimation algorithms fail to fully utilize prior information from color images, resulting in poor quality depth estimation images, especially when encountering scenarios not encountered during training, and face difficulties in model upgrades post-deployment.

Innovation Solution

A depth estimation method that utilizes a first depth estimation model to obtain an intermediate depth image from a color image, followed by a second model using the color and intermediate depth images to enhance the target depth image, incorporating common edge extraction and feature vector processing, and enables model updates through gradient information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If monocular images are used for depth estimation, then device complexity is reduced, but measurement precision deteriorates due to ambiguous depth information

Engineering Contradiction:
Improvedevice complexityVSAvoiddepth estimation precision
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent introduces a depth estimation model as an intermediary component that processes monocular images and generates depth maps. This model acts as a mediator between the simple monocular input and the required depth information, enabling depth estimation from single images while maintaining reasonable accuracy through learned features and contextual reasoning

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If depth estimation models process high-resolution images, then measurement precision improves, but processing time increases

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

Solution Approach 1:

The patent segments the depth estimation process into multiple stages: initial depth map generation, error map creation, and refinement processing. By dividing the task into sequential steps that operate on different resolutions and focus on different aspects (gross depth structure vs. fine details), the system achieves high precision without requiring the entire process to operate at full resolution simultaneously

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies partial action by generating depth maps at different resolution levels and applying refinement selectively. The refinement model processes only the necessary portions of the image at high resolution, applying excessive computation only where needed to achieve precise depth estimation while maintaining overall efficiency

Inventive Principle:
Principle #16Partial or excessive action

3Measurement precision

If refinement models are applied to correct depth errors, then measurement precision improves, but device complexity increases

Engineering Contradiction:
Improvedepth estimation precisionVSAvoidmodel complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces an error map as an intermediary representation that captures depth estimation errors without requiring direct complex interactions between multiple refinement models. This error map serves as a mediator that guides the refinement process, enabling precise corrections while maintaining manageable system complexity through a structured intermediate representation

Inventive Principle:
Principle #24Intermediary (Mediator)

Data Source

PatentEP4379660B1Depth estimation method, method for training depth estimation model, and apparatus and system thereof
Publication Date: 2026.04.29 YINWANG INTELLIGENT TECHNOLOGIES CO LTD
  • EP4379660B1 patent drawingFigure 1
  • EP4379660B1 patent drawingFigure 2~3
  • EP4379660B1 patent drawingFigure 4

AI summary

This application provides a depth estimation method and apparatus, a method and an apparatus for training a depth estimation model, and a system. The method includes: obtaining a first color image; inputting the first color image into a first depth estimation model, to obtain a first intermediate depth image; and inputting the first color image and the first intermediate depth image into a second depth estimation model, to obtain a first target depth image, where the second depth estimation model is obtained through training based on a color image and a target depth image corresponding to the color image, and the first depth estimation model is obtained through training based on the color image and an intermediate depth image corresponding to the color image. According to the technical solution of this application, the intermediate depth image corresponding to the color image is obtained by using the color image, and the target depth image corresponding to the color image is obtained based on the color image and the intermediate depth image corresponding to the color image. This can fully utilize prior information, namely, the color image, and help improve precision of a target depth estimation image.