Neural Network Object Detection Distance Estimation

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional object detection technologies using monocular cameras struggle to accurately estimate the distance to objects when assumptions do not apply, leading to incorrect distance calculations.

Innovation Solution

An object detecting apparatus and method that utilizes a neural network to simultaneously estimate the posture and distance of objects in a captured image, incorporating non-linear processing and a detecting unit to identify object-candidate regions and output object information, including distance, using a neural network trained for both tasks.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Device complexity

If conventional object detection technologies use monocular cameras with assumptions to estimate distance, then the detection process is simple, but the distance estimation accuracy deteriorates when assumptions do not apply

Engineering Contradiction:
Improvedetection process complexityVSAvoiddistance estimation accuracy
Core Design Contradiction:
Device complexityVSMeasurement precision

Solution Approach 1:

The patent combines object detection and distance estimation into a unified neural network model that processes the captured image simultaneously for both tasks. The detecting unit identifies object-candidate regions while the non-linear processing unit estimates distance using the same image data, merging two previously separate functions into one integrated system that improves distance accuracy without proportionally increasing complexity

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The neural network is designed to perform multiple functions: detecting objects, estimating their posture, and calculating distance simultaneously. This multi-functional approach allows the system to leverage shared features and computations across tasks, improving distance estimation accuracy while maintaining reasonable system complexity through efficient resource utilization

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

2Device complexity

If conventional technologies detect objects and estimate distance separately, then the detection algorithm is simple, but the overall detection accuracy deteriorates

Engineering Contradiction:
Improvealgorithm complexityVSAvoiddetection accuracy
Core Design Contradiction:
Device complexityVSReliability

Solution Approach 1:

The patent merges object detection and distance estimation into a single integrated neural network processing pipeline. The detecting unit and non-linear processing unit work together to simultaneously identify objects and calculate their distance from the captured image, ensuring that both tasks benefit from shared feature extraction and coordinated optimization, thereby improving overall detection reliability

Inventive Principle:
Principle #5Merging (Combining)

Solution Approach 2:

The system employs feedback mechanisms where the output of the detecting unit informs the non-linear processing unit, and the distance estimation results can be used to refine object detection. This iterative feedback loop between detection and distance estimation allows the system to continuously improve accuracy by leveraging information from both tasks

Inventive Principle:
Principle #23Feedback

Data Source

PatentEP3418943B1Object detecting apparatus, object detecting method, and computer-readable medium
Publication Date: 2023.11.29 KK TOSHIBA
  • EP3418943B1 patent drawingFigure 1~2
  • EP3418943B1 patent drawingFigure 3
  • EP3418943B1 patent drawingFigure 4

AI summary

According to an arrangement, an object detecting apparatus (1) includes a detecting unit (10; 12) and a non-linear processing unit (10; 13). The detecting unit (10; 12) is configured to detect one or more object-candidate regions from a captured image. The non-linear processing unit (10; 13) is configured to input the entire captured image or a part of the captured image at least including the object-candidate region to a neural network having been trained to estimate the posture of an object in the object-candidate region and the distance to the object simultaneously, and output object information at least including information on the distance to the object, using an output from the neural network.