Vehicle Free-Space Estimation Using Depth-Segmentation Grids

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional methods for free-space estimation in autonomous driving, such as depth image generation and pixel-by-pixel semantic segmentation, are susceptible to noise and fail to accurately handle non-flat objects, leading to inaccuracies in determining drivable areas around vehicles.

Innovation Solution

A method that combines depth information with semantic segmentation data to estimate free-space on a grid representing the ground plane, using aggregated confidence scores to improve accuracy and robustness by weighing reference points and handling outliers, allowing for real-time processing with reduced computational resources.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If depth image generation is used for free-space estimation, then the method can determine free-space from flat surfaces, but it becomes susceptible to noise in flat areas

Engineering Contradiction:
Improvefree-space estimation accuracyVSAvoidnoise susceptibility
Core Design Contradiction:
Measurement precisionVSReliability

Solution Approach 1:

The patent combines depth information with semantic segmentation information into a unified free-space estimation framework. By merging these two data sources, the system leverages the strengths of both approaches while compensating for their individual weaknesses, thereby improving reliability without sacrificing precision.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent introduces an intermediary aggregation process that combines confidence scores from depth data and segmentation data. This intermediary step acts as a mediator that filters and reconciles the two data sources, reducing noise susceptibility while maintaining estimation accuracy.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Measurement precision

If pixel-by-pixel semantic segmentation is performed to classify road vs non-road, then free-space can be determined as road areas, but it fails to accurately handle non-flat objects

Engineering Contradiction:
Improveroad classification accuracyVSAvoidhandling of non-flat objects
Core Design Contradiction:
Measurement precisionVSAdaptability or versatility

Solution Approach 1:

The patent transitions from two-dimensional pixel-by-pixel classification to a three-dimensional grid-based approach that incorporates depth information. This dimensional change enables the system to handle non-flat objects effectively by considering their spatial extent and depth, while maintaining road classification accuracy through the integration of semantic segmentation.

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

3Measurement precision

If conventional free-space estimation methods are used, then computational resources can be managed, but accuracy and robustness are insufficient

Engineering Contradiction:
Improvefree-space estimation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the environment into a grid structure and processes each cell independently based on aggregated confidence scores from reference points. This segmentation approach maintains computational efficiency by dividing the problem into manageable units while improving accuracy through comprehensive data aggregation within each cell.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4468257A1A method and device for determining free-space
Publication Date: 2024.11.27 ZENSEACT AB
  • EP4468257A1 patent drawingFigure 1
  • EP4468257A1 patent drawingFigure 2
  • EP4468257A1 patent drawingFigure 3

AI summary

The present invention is related to a computer-implemented method (100) for determining free-space in a surrounding environment of a vehicle equipped with an automated driving system, ADS. The method (100) comprises: obtaining (S102), from a camera of the vehicle, an image depicting the surrounding environment of the vehicle; determining (S104), for each reference point of a plurality of reference points in the image, segmentation data indicative of a confidence score of the reference point belonging to free-space; determining (S106), for each reference point of the plurality of reference points, depth data indicative of a distance of each reference point relative the camera; assigning (S110) each reference point of the plurality of reference points to a cell (406) of a plurality of cells in a grid (408), based on the depth data associated with each reference point of the plurality of reference points, wherein the grid (408) corresponds to an estimated ground plane of the surrounding environment; and assigning (S118), to each cell of the grid, an aggregated confidence score, based on the segmentation data associated with each reference point of the respective cell, thereby determining a free-space in the surrounding environment of the vehicle as the cells having aggregated confidence scores indicative of belonging to free-space.