Vehicle Object Representation via Heading-Aligned Sensor Data Aggregation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current vehicle sensor systems face challenges in accurately representing and classifying detected objects in traffic environments, particularly due to sparse detection data and varying viewing angles, which complicates efficient object representation and classification.

Innovation Solution

A method that transforms vehicle sensor data into a representation coordinate system aligned with the object's heading, aggregates detections over cycles, and incorporates Doppler and signal strength data to create a robust and efficient object representation, enabling effective classification using neural networks like CNNs.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If point models and polygon models are used to represent tracked objects, then object representation accuracy is improved, but computational complexity increases

Engineering Contradiction:
Improveobject representation accuracyVSAvoidcomputational complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the object representation into discrete detection points rather than using continuous geometric models. Each detection point retains its measured properties (position, velocity, signal strength) and is processed independently, avoiding the computational burden of fitting and manipulating complex geometric models while preserving measurement accuracy.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transforms the problem from spatial geometry representation to a multi-dimensional feature space by incorporating detection properties (position, velocity, signal strength, Doppler shift) as separate dimensions. This allows accurate object representation through attribute-based classification rather than geometric modeling.

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

2Measurement precision

If sensor data from multiple cycles are aggregated to improve object representation, then detection accuracy is improved, but data processing complexity increases due to varying viewing angles

Engineering Contradiction:
Improvedetection accuracyVSAvoiddata processing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

Instead of transforming sensor data into a fixed coordinate system and losing angular information, the patent inverts the approach by using the detection data itself (including angular variations) as the representation basis. The varying viewing angles are preserved as valuable information rather than being normalized away, enabling classification based on detection properties across different angles.

Inventive Principle:
Principle #13The other way round (Inversion)

Solution Approach 2:

The patent changes the representation parameters from fixed geometric coordinates to dynamic detection properties (position, velocity, signal strength, Doppler shift) that naturally adapt to varying viewing angles. This allows aggregation of multi-cycle data without requiring complex coordinate transformations, as the detection properties are inherently frame-invariant.

Inventive Principle:
Principle #35Parameter changes

3Measurement precision

If Doppler data and signal strength data are incorporated into object representation, then object classification accuracy is improved, but processing requirements increase

Engineering Contradiction:
Improveobject classification accuracyVSAvoidprocessing requirements
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent merges multiple detection properties (position, velocity, signal strength, Doppler shift) into a unified object representation framework. By combining these different types of data in a single processing pipeline, the system achieves accurate classification without requiring separate processing chains for each data type, thereby reducing overall processing requirements.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal detection representation that serves multiple functions simultaneously: object tracking, classification, and angular variation handling. The same set of detection properties (including Doppler and signal strength) is used for all these purposes, eliminating the need for specialized processing for each function and reducing total processing requirements.

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

Applied Scientific Principles

This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.

Function Achieved in This Case

This approach allows for accurate and efficient representation and classification of objects, compensating for sparse data and varying viewing angles, improving the applicability to diverse traffic scenarios and enhancing decision-making in autonomous driving systems.

Implementation Method 1

vehicle environment detection systems, such as for example camera systems, radio detection and ranging (radar) systems

Methodology Applied
Scientific EffectRadar: Radar

Implementation Method 2

Detections comprising detection coordinates and Doppler data

Methodology Applied
Scientific EffectDoppler effect: Doppler Effect

Data Source

PatentEP3588136B1Object representation and classification based on vehicle sensor detections
Publication Date: 2021.03.03 VEONEER SWEDEN AB
  • EP3588136B1 patent drawingFigure 1
  • EP3588136B1 patent drawingFigure 2~4
  • EP3588136B1 patent drawingFigure 5

AI summary

A method for representing an object using vehicle sensor (110) data. The method comprising, for each cycle in a plurality of cycles; obtaining (S1) detections related to the object (150, 160), the detections comprising detection coordinates, the object being associated with a heading (Vg1, Vg2), transforming (S2) the detections into a representation coordinate system, the representation coordinate system having an orientation determined based on the heading (Vg1, Vg2), aggregating (S4) transformed detection coordinates with transformed detection coordinates from previous cycles, and representing (S5) the object by the aggregated detections.