Monocular Object Localization via Dynamic Ground Plane Estimation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Three-dimensional localization of moving objects in a video stream using a single camera is challenging due to instability in feature tracks for objects like cars and high localization errors from triangulating object bounding boxes against a fixed ground plane, often requiring expensive inference mechanisms.

Innovation Solution

The method estimates a ground plane in a video frame using monocular structure-from-motion (SFM) information and dense feature tracking, computing object pose through intensity alignment and epipolar constraints, which reduces reprojection errors and stabilizes localization by combining cues from SFM, dense inter-frame stereo, and object detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If sparse feature points are used for localization, then computational complexity is reduced, but localization accuracy and stability deteriorate because stable feature tracks are hard to establish on objects like cars

Engineering Contradiction:
Improvecomputational complexityVSAvoidlocalization accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent changes the parameter of feature density from sparse to dense. Instead of using a limited number of feature points, the system employs dense feature tracking across the entire object surface, which provides sufficient stable tracks for accurate localization while maintaining computational feasibility through efficient processing algorithms

Inventive Principle:
Principle #35Parameter changes

Solution Approach 2:

The patent transitions from two-dimensional image plane features to three-dimensional spatial localization. By using dense feature tracking in 3D space and incorporating ground plane constraints, the system achieves accurate 3D localization of moving objects without requiring expensive multi-camera setups

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

2Ease of operation

If object bounding boxes are triangulated against a fixed ground plane, then localization can be achieved, but high localization errors occur due to the fixed plane assumption not adapting to actual ground variations

Engineering Contradiction:
Improvelocalization feasibilityVSAvoidlocalization accuracy
Core Design Contradiction:
Ease of operationVSMeasurement precision

Solution Approach 1:

The patent makes the ground plane dynamic rather than fixed. The system estimates and adapts the ground plane parameters based on detected objects and dense feature tracking, allowing the ground plane to adjust to actual terrain variations and moving object positions, thereby eliminating systematic localization errors

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The system uses feedback from dense feature tracking and object detection to continuously refine ground plane estimates. This closed-loop approach allows the ground plane model to adapt to changing scenes and object positions, improving localization accuracy over time

Inventive Principle:
Principle #23Feedback

3Measurement precision

If expensive inference mechanisms are used, then localization accuracy can be improved, but computational cost and processing time increase significantly

Engineering Contradiction:
Improvelocalization accuracyVSAvoidcomputational cost
Core Design Contradiction:
Measurement precisionVSUse of energy by moving object

Solution Approach 1:

The patent replaces expensive, complex inference mechanisms with simpler, more efficient algorithms. By using dense feature tracking with ground plane constraints and optimized computational approaches, the system achieves high localization accuracy at a fraction of the computational cost of traditional methods

Inventive Principle:
Principle #27Cheap short-living objects (Disposable)

Data Source

PatentEP3114833B1High accuracy monocular moving object localization
Publication Date: 2019.10.16 NEC CORP
  • EP3114833B1 patent drawingFigure 1~2
  • EP3114833B1 patent drawingFigure 3
  • EP3114833B1 patent drawingFigure 4

AI summary

Methods and systems for moving object localization include estimating a ground plane in a video frame based on a detected object within the video frame and monocular structure-from-motion (SFM) information; computing object pose for objects in the frame based on the SFM information using dense feature tracking; and determining a three-dimensional location for the detected object based on the estimated ground plane and the computed object pose.