3D Camera Curb Detection Without LiDAR or Flat-World Assumptions

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing curb detection methods in advanced driver assistance systems face inaccuracies and high computational demands, particularly in complex environments, and rely heavily on costly and inefficient LiDAR systems, and existing camera systems, and existing camera-based methods for detecting curbs are prone to inaccuracies due to simplifications like flat-world assumptions.

Innovation Solution

A computer-implemented method using camera images to determine curb location and depth, employing neural networks for object detection and segmentation, transforming pixel coordinates, and applying edge of the curb and assigned depth information into a three dimensional reference coordinate system, determining the edge of the curb based on transformed image pixels and assigned depth information.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If LiDAR sensors are used for curb detection, then measurement precision is improved, but device complexity and cost increase

Engineering Contradiction:
Improvecurb detection precisionVSAvoidsensor system complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces LiDAR sensors with a camera-based vision system for curb detection. The camera captures images that are processed through neural networks to detect curbs, eliminating the need for complex LiDAR hardware while achieving comparable or superior detection precision through image processing and depth estimation algorithms.

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

Solution Approach 2:

The patent creates a virtual 3D representation of the curb from 2D camera images by estimating depth information and transforming image coordinates into a three-dimensional reference coordinate system. This virtual model replicates the physical curb's spatial characteristics without requiring physical LiDAR scanning.

Inventive Principle:
Principle #26Copying

2Measurement precision

If LiDAR sensors are used for curb detection, then measurement precision is improved, but processing speed decreases

Engineering Contradiction:
Improvecurb detection precisionVSAvoidprocessing speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent performs preliminary depth estimation and curb detection directly on the 2D camera image before transforming to 3D space. The neural network processes the image to identify potential curb regions and estimate depth, then only transforms the detected curb coordinates to 3D, avoiding the computationally intensive processing of entire point clouds that LiDAR requires.

Inventive Principle:
Principle #10Preliminary action

3Device complexity

If flat-world assumption is used for camera-based curb detection, then device complexity is reduced, but measurement precision deteriorates

Engineering Contradiction:
Improvedetection method simplicityVSAvoidcurb position accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent transitions from 2D image coordinates to a three-dimensional reference coordinate system by estimating depth information for pixels corresponding to detected curbs. This dimensional transformation allows accurate representation of curb position in 3D space without requiring the simplifying flat-world assumption, maintaining both simplicity and precision.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4660963A1Image based curb detection
Publication Date: 2025.12.10 AUMOVIO AUTONOMOUS MOBILITY GERMANY GMBH
  • EP4660963A1 patent drawingFigure 1
  • EP4660963A1 patent drawingFigure 2a~2b
  • EP4660963A1 patent drawingFigure 3

AI summary

The invention is concerned with a computer-implemented method for determining a curb (C), a computer program and a computer-readable (storage) medium. The method comprises the steps of: - receiving an image (I(C)) from a camera showing at least one curb (C); - determining a location of the curb (C) in the image and image pixels p(C) corresponding to the curb (C); - determining a depth map (d(I)) at least for an image section sowing the curb (C); - assigning depth information (d) of the depth map (d(I)) to each image pixel corresponding to the curb (p(C,d)); - transforming the image pixel (p(C,d)) corresponding to the curb (C) and assigned depth information (d) into a three dimensional reference coordinate system (p'(C,d)); and - determining an edge (Ed) of the curb (C) based on the transformed image pixel (p(C,d)) corresponding to the curb (C) and assigned depth information (d).