Vehicle Camera Object Detection Using Planar Homography Segmentation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current vehicle detection systems based on classification methods struggle with new vehicle designs and changing structures, leading to reduced performance, and monocular systems face challenges in accurately triangulating objects near the epipole and estimating dynamic objects' speed due to unknown relative speed.

Innovation Solution

A method for detecting objects from a sequence of images using a vehicle camera, involving the steps of recording images, determining corresponding features, calculating homographies for neighboring features, segmenting features based on levels, and detecting objects by assigning features to levels, allowing for generic object recognition and improved estimation of dynamic objects' speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If classification-based methods are used for vehicle recognition, then known vehicle designs can be recognized, but new vehicle designs and changing body structures lead to significantly reduced system performance

Engineering Contradiction:
Improvevehicle recognition accuracyVSAvoidadaptability to new vehicle designs
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent changes the fundamental parameters of object representation from classification categories to geometric homography parameters. By representing objects through their planar geometry and motion characteristics rather than predefined classes, the system can adapt to any vehicle design without retraining, resolving the contradiction between recognition accuracy for known vehicles and adaptability to new designs

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The homography-based approach creates a universal detection framework that works across all vehicle types and designs. The method detects planar structures and tracks their motion universally, making the system multi-functional for recognizing any object with planar surfaces regardless of its specific class or design characteristics

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

2Measurement precision

If monocular triangulation is used for object detection, then object positions can be estimated, but objects near the epipole cannot be triangulated precisely and smallest errors in egomotion become noticeable

Engineering Contradiction:
Improveobject position triangulation accuracyVSAvoiddetection difficulty near epipole
Core Design Contradiction:
Measurement precisionVSDifficulty of detecting and measuring

Solution Approach 1:

The patent extracts and removes objects near the epipole from the problematic monocular triangulation process by detecting them as separate planar structures through homography. This extracted approach bypasses the epipolar geometry limitations that cause precision loss near the epipole, allowing accurate detection in previously problematic regions

Inventive Principle:
Principle #2Taking out (Extraction)

Solution Approach 2:

The patent introduces homography computation as an intermediary method between feature matching and object localization. This intermediary approach uses planar geometry constraints to mediate the positioning problem, providing accurate results near the epipole where direct triangulation fails

Inventive Principle:
Principle #24Intermediary (Mediator)

3Loss of information

If homography calculation is performed for all corresponding features, then complete scene understanding can be achieved, but computational complexity increases

Engineering Contradiction:
Improvescene interpretation completenessVSAvoidcomputational complexity
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent segments the image into multiple planar regions and calculates homographies separately for each region rather than for the entire image. This segmentation reduces computational complexity by processing smaller, localized areas while maintaining complete scene understanding through the aggregation of regional homographies

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies homography calculation selectively to regions containing planar structures rather than uniformly across the entire image. This partial action approach performs computations only where needed (excessive in relevant regions, absent in irrelevant regions), optimizing the balance between scene understanding completeness and computational complexity

Inventive Principle:
Principle #16Partial or excessive action

Data Source

PatentEP3520024B1Detection and validation of objects from sequential images of a camera using homograpies
Publication Date: 2024.09.11 CONTINENTAL AUTONOMOUS MOBILITY GERMANY GMBH
  • EP3520024B1 patent drawingFigure 1~2
  • EP3520024B1 patent drawingFigure 3a~4
  • EP3520024B1 patent drawingFigure 5a~5d

AI summary

The invention relates to a method and to a device for identifying objects from images from a camera and can be used in particular in camera-based driver assistance systems. The method for detecting objects from a series of images from a vehicle camera comprises the following steps: a) capturing a series of images by means of the vehicle camera, b) determining corresponding features in two consecutive images, d) determining a plurality of planes in space by associating adjacent corresponding features with a plane in space, f) detecting objects while taking into account the planes determined in step d). According to a preferred embodiment, the method comprises the following step: c) computing homographs for adjacent corresponding features for determining the plurality of planes; and/or d2) associating adjacent corresponding features with a bottom plane, a back wall plane, or a side wall plane. Because the method according to the invention can be designed to be very fast, generic object detection or scene interpretation is possible for nearly the whole image in real time.