3D Camera Object Detection Using Reference Data and Model Objects

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing object detection methods using 3D cameras face challenges in processing extensive measurement data efficiently, requiring high computing effort, while needing to operate in real-time and maintain high reliability, especially in variable lighting conditions and safety-critical applications like autonomous vehicles.

Innovation Solution

A method that utilizes both object detection data and reference detection data, with the latter acquired under controlled lighting conditions, to normalize and compensate for ambient light variations, and employs geometrically defined model objects to simplify data processing and enhance detection reliability, using a combination of distance and intensity values from the 3D camera.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If extensive measurement data from 3D camera is processed in detail, then detection reliability is improved, but processing time increases and real-time performance deteriorates

Engineering Contradiction:
Improvedetection reliabilityVSAvoidprocessing time
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The patent segments the measurement data processing into two distinct phases: a reference measurement phase that captures comprehensive data for building a detailed environmental model, and a subsequent object detection phase that uses this pre-built model for faster comparison-based detection. This segmentation allows detailed processing to occur only once during reference capture, while subsequent detections use simplified comparison operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs preliminary action by capturing reference measurements of the environment before actual object detection begins. This reference data, including depth maps and normal maps under known lighting conditions, is pre-processed and stored as a baseline. During actual detection, the system compares current measurements against this pre-established reference, avoiding the need to process full measurement data from scratch each time.

Inventive Principle:
Principle #10Preliminary action

2Reliability

If comprehensive measurement data is processed to ensure high detection reliability, then computing effort increases

Engineering Contradiction:
Improvedetection reliabilityVSAvoidcomputing effort
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The patent performs preliminary action by capturing reference measurements of the environment before actual object detection begins. This reference data, including depth maps and normal maps under known lighting conditions, is pre-processed and stored as a baseline. During actual detection, the system compares current measurements against this pre-established reference, avoiding the need to process full measurement data from scratch each time.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent creates simplified copies or representations of the environmental data in the form of depth maps and normal maps. These map representations serve as computationally efficient proxies for the full 3D measurement data, allowing the system to perform complex comparisons and object detection algorithms using the streamlined map data rather than raw sensor measurements.

Inventive Principle:
Principle #26Copying

3Productivity

If object detection is performed quickly in real-time, then processing speed is improved, but detection reliability may deteriorate

Engineering Contradiction:
Improveprocessing speedVSAvoiddetection reliability
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent segments the measurement data processing into two distinct phases: a reference measurement phase that captures comprehensive data for building a detailed environmental model, and a subsequent object detection phase that uses this pre-built model for faster comparison-based detection. This segmentation allows detailed processing to occur only once during reference capture, while subsequent detections use simplified comparison operations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent creates simplified copies or representations of the environmental data in the form of depth maps and normal maps. These map representations serve as computationally efficient proxies for the full 3D measurement data, allowing the system to perform complex comparisons and object detection algorithms using the streamlined map data rather than raw sensor measurements.

Inventive Principle:
Principle #26Copying

4Measurement precision

If variable lighting conditions are compensated for using reference measurements, then detection accuracy is improved, but measurement and processing time increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidmeasurement time
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The patent performs preliminary action by capturing reference measurements of the environment before actual object detection begins. This reference data, including depth maps and normal maps under known lighting conditions, is pre-processed and stored as a baseline. During actual detection, the system compares current measurements against this pre-established reference, avoiding the need to process full measurement data from scratch each time.

Inventive Principle:
Principle #10Preliminary action

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 enables fast and reliable object detection, improving processing efficiency and maintaining high detection accuracy even in changing environments, thereby reducing the risk of incorrect controls and collisions in autonomous vehicle applications.

Implementation Method 1

a 3D camera, preferably a TOF (time of flight) camera

Methodology Applied
Scientific EffectTime of flight: Time of Flight

Implementation Method 2

an illumination unit and an image sensor, with the measured values preferably being determined separately for individual pixels of the image sensor

Methodology Applied
Scientific EffectLight reflection: Reflection

Data Source

PatentEP3663800B1Method for detecting objects using a 3D camera
Publication Date: 2021.01.27 SICK AG
  • EP3663800B1 patent drawingFigure 1
  • EP3663800B1 patent drawingFigure 2
  • EP3663800B1 patent drawingFigure 3

AI summary

Object detection is performed using an optical sensor comprising at least one 3D camera with an illumination unit and an image sensor, wherein the optical sensor is adapted to detect objects in a detection area of ​​the 3D camera based on measured values ​​of the detection area, wherein the measured values ​​are determined separately for individual pixels of the image sensor.The method comprises at least the following steps: determining object detection data comprising at least measured values ​​of the detection range, which are determined when objects are present in the detection range and the detection range is preferably illuminated by radiation emitted by the lighting unit; determining reference detection data comprising at least measured values ​​of the detection range, which are determined when no object is present in the detection range and/or the detection range is not illuminated by radiation emitted by the lighting unit; and determining model objects based on the object detection data and the reference detection data in order to detect objects in the detection range, each model object representing each object in the detection range.