Reflective Surface Object Detection With False-Positive Filtering

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Automated driving systems face challenges in accurately distinguishing between real and reflected objects in environments with reflective surfaces, leading to false positive detections that can compromise safety.

Innovation Solution

A system and method utilizing a neural network to differentiate between real and reflected objects based on geometric coordinates, classification scores, uncertainty values, reflection scores, and similarity scores, enhanced by V2V communication for improved accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If conventional object detection methods are used on reflective surfaces, then detection speed can be maintained, but detection accuracy deteriorates due to false positives and false negatives caused by light reflections

Engineering Contradiction:
Improveobject detection accuracyVSAvoidfalse positives and false negatives from light reflections
Core Design Contradiction:
Measurement precisionVSObject-affected harmful factors

Solution Approach 1:

The patent introduces a light diffuser as an intermediary element between the light source and the reflective surface. This diffuser scatters the light before it reaches the object, creating multiple reflected image paths that can be detected. The diffuser acts as a mediator that transforms direct reflected light into distributed light patterns, enabling more reliable object detection on reflective surfaces by reducing the impact of specular reflections.

Inventive Principle:
Principle #24Intermediary (Mediator)

Solution Approach 2:

The patent transitions from direct 2D image capture to 3D depth information capture using time-of-flight measurement. By measuring the time for light to travel to and from the object, the system obtains depth data that is independent of surface reflectivity. This dimensional change from intensity-based detection to time-based detection eliminates the harmful effect of light reflections on detection accuracy.

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

2Measurement precision

If multiple reflected images are captured to improve detection accuracy, then object detection precision improves, but the complexity of processing multiple images increases

Engineering Contradiction:
Improveobject detection precisionVSAvoidprocessing complexity of multiple reflected images
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent replaces complex mechanical image processing with optical time measurement. Instead of capturing and processing multiple reflected images to determine object presence, the system directly measures the time of flight of light. This substitution of mechanical/image processing with temporal measurement simplifies the system while maintaining high detection precision.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

Solution Approach 2:

The system uses the light reflections that would normally be harmful as the very mechanism for detection. By measuring the time for light to reflect off the object and return, the system turns the reflected light itself into the detection signal, eliminating the need for separate processing of multiple images while maintaining accuracy.

Inventive Principle:
Principle #25Self-service

3Measurement precision

If time-of-flight measurement is used to detect objects on reflective surfaces, then detection accuracy improves by ignoring surface reflectivity, but additional measurement infrastructure is required

Engineering Contradiction:
Improvedetection accuracy independent of surface reflectivityVSAvoidadditional measurement infrastructure for time-of-flight
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent combines the light source, diffuser, and time-of-flight sensor into an integrated detection system. The light source and sensor are positioned in close proximity, and the diffuser is integrated into the optical path. This merging of components reduces the overall system complexity despite adding time-of-flight capability, as the components work together in a unified optical arrangement rather than separate systems.

Inventive Principle:
Principle #5Merging (Combining)

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

Reduces false-positive detections by effectively distinguishing between real and reflected objects, enhancing the safety and reliability of automated driving systems in reflective environments.

Implementation Method 1

a light source positioned close to the image sensor in proximity to reflect light off the object

Methodology Applied
Scientific EffectReflection: Reflection

Implementation Method 2

an imaging lens positioned between the light source and the object

Methodology Applied
Scientific EffectRefraction: Refraction

Data Source

PatentEP4272114B1Improved object detection on reflective surfaces
Publication Date: 2026.05.20 YINWANG INTELLIGENT TECHNOLOGIES CO LTD
  • EP4272114B1 patent drawingFigure 1
  • EP4272114B1 patent drawingFigure 2
  • EP4272114B1 patent drawingFigure 3

AI summary

A system (220) and method (300,400,500,600,700) is provided for object detection within a reflective surface (101,760). The system (220) comprises a processor (225,430) configured to receive at least one input processor frame (710,410) as sequence of input image frames (710,410) and to detect a reflective surface (101,760) in the sequence of input image frames (710,410). The processor is configured to perform object detection within the reflective surface (101,760) and differentiate objects detected in the reflective surface (101,760) as real objects or reflected objects, thereby reducing false-positive detections, wherein the differentiation is based on at least one of geometric coordinates of a respective detection bounding box, a classification score, an associated uncertainty value, a reflection score or a similarity score.