Vehicle Motion Modeling With Segmentation and Geometric Constraints

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing methods struggle to accurately detect and model the motion of single moving objects in the surroundings of a motor vehicle, particularly when they are close to each other, due to challenges in training neural networks with unknown objects and dynamic scenarios.

Innovation Solution

A neural network-based method that incorporates prior information as 'soft constraints' and uses a multi-task learning approach with loss functions and geometric constraints to enhance detection and segmentation, including 2D and 3D bounding boxes, spatial transformations, and optical flow analysis.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If a neural network is trained to detect moving objects, then detection capability is improved, but the system fails to handle unknown objects (e.g., kangaroos) that are not represented in training data

Engineering Contradiction:
Improvedetection capabilityVSAvoidhandling of unknown objects
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The patent segments the detection task into multiple specialized neural networks, each trained to detect specific object categories (pedestrians, vehicles, animals, etc.). This segmentation allows the system to maintain high reliability for known objects while providing a structured framework to handle unknown objects through fallback mechanisms or novel object detection modules.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent implements a universal detection framework that combines multiple specialized detectors with a general novel object detection capability. The system can adaptively switch between specialized detectors for known objects and general detection mechanisms for unknown objects, achieving both reliability and versatility.

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

2Measurement precision

If traditional object detection methods are used, then static objects can be detected from HD maps, but moving objects are harder to detect and form highly dynamic scenarios

Engineering Contradiction:
Improvestatic object detectionVSAvoidmoving object detection
Core Design Contradiction:
Measurement precisionVSDifficulty of detecting and measuring

Solution Approach 1:

The patent employs dynamic detection strategies that adapt to the motion characteristics of objects. Multiple neural networks are trained with different focal losses optimized for detecting moving objects with varying speeds and trajectories. The system dynamically adjusts detection parameters based on object motion patterns, improving detection difficulty handling while maintaining precision.

Inventive Principle:
Principle #15Dynamics

Solution Approach 2:

The patent implements continuous tracking and detection mechanisms that maintain useful action across frames. By using temporal consistency and frame-by-frame analysis with specialized motion detection networks, the system maintains continuous detection capability for moving objects, overcoming the difficulty of detecting highly dynamic scenarios.

Inventive Principle:
Principle #20Continuity of useful action

3Measurement precision

If multiple neural networks are used for different object types, then detection accuracy for specific objects is improved, but device complexity increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidsystem complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent merges multiple specialized neural networks into a unified detection system that shares common feature extraction layers. The architecture combines pedestrian, vehicle, and animal detectors with shared backbone networks, reducing redundant computations and lowering overall system complexity while maintaining high detection accuracy for each object type.

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The patent creates a universal detection framework where a single system performs multiple detection functions. The multi-functional architecture uses shared resources and coordinated neural networks to detect various object types, reducing device complexity compared to completely separate detection systems while preserving specialized detection accuracy.

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

Data Source

PatentEP4226323B1Method for determining a motion model of an object in the surroundings of a motor vehicle, computer program product, computer-readable storage medium, as well as assistance system
Publication Date: 2026.04.22 CONNAUGHT ELECTRONICS
  • EP4226323B1 patent drawingFigure 1
  • EP4226323B1 patent drawingFigure 2
  • EP4226323B1 patent drawingFigure 3

AI summary

The invention relates to a method for determining a motion model (3) of an object (4) by an assistance system (2), the method comprising the steps: - capturing an image (20, 21) of the surroundings (5) with the moving object (4) by a capturing device (6); - encoding the image (20, 21) by a feature extraction module (9) of a neural network (8) of an electronic computing device (7); - decoding the encoded image (20, 21) by an object segmentation module (10) and generating a first loss function (22); - decoding the at least one encoded image (20, 21) by a bounding box estimation module (11) and generating a second loss function (23); - decoding the second loss function (23) depending on the decoding of the image (20, 21) by a motion decoding module (12) and generating a third loss function (24); and - determining the motion model (3) depending on the first loss function (22) and the third loss function (24). Further the invention relates to a computer program product, a computer-readable storage medium, as well as an assistance system (2).