Camera-Based Neural Networks for Object Identification in Water

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing systems fail to reliably detect and identify objects partially or fully submerged in water, such as whales or debris, due to limitations in sonar, radar, and lidar technologies, which can lead to collisions with boats.

Innovation Solution

A system using cameras and artificial neural networks to generate and process images of water areas, allowing detection and identification of submerged objects by comparing extracted features with predefined sets, and optionally incorporating inertial measurement for compensation and horizon detection.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If sonar-based technologies are used to detect immersed objects, then detection capability for underwater objects is improved, but detection capability for floating objects is worsened

Engineering Contradiction:
Improvedetection capabilityVSAvoiddetection coverage
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system combines multiple detection technologies (sonar for underwater objects, radar/lidar for floating objects, and optical cameras for identification) into a single integrated detection system, enabling it to detect both immersed and floating objects effectively

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

2Length of stationary object

If radar systems are used to detect floating objects, then detection range is improved, but identification capability is worsened

Engineering Contradiction:
Improvedetection rangeVSAvoididentification capability
Core Design Contradiction:
Length of stationary objectVSMeasurement precision

Solution Approach 1:

The system uses radar as an intermediary detection layer that provides early warning of floating objects at long ranges, then directs optical cameras to focus on identified targets for precise classification and identification at closer distances

Inventive Principle:
Principle #24Intermediary (Mediator)

3Measurement precision

If lidar systems are used to detect floating objects, then identification capability is improved, but detection range is worsened

Engineering Contradiction:
Improveidentification capabilityVSAvoiddetection range
Core Design Contradiction:
Measurement precisionVSLength of stationary object

Solution Approach 1:

The system applies lidar selectively only to regions of interest identified by radar or sonar, rather than scanning the entire water area, thereby achieving high identification precision while maintaining efficient use of resources and extended effective detection range

Inventive Principle:
Principle #16Partial or excessive action

4Reliability

If signals are used to detect floating objects, then detection capability is improved, but signal reliability is worsened

Engineering Contradiction:
Improvedetection capabilityVSAvoidsignal integrity
Core Design Contradiction:
ReliabilityVSLoss of information

Solution Approach 1:

The system uses optical cameras to capture visual images of detected objects, creating a visual copy that can be analyzed and identified without relying on the potentially lossy or modified signal reflections from the object

Inventive Principle:
Principle #26Copying

Data Source

PatentEP3881220B1System and method for identifying an object in water
Publication Date: 2025.10.29 SEA AI GMBH
  • EP3881220B1 patent drawingFigure 1~3
  • EP3881220B1 patent drawingFigure 4~5
  • EP3881220B1 patent drawingFigure 6

AI summary

The invention relates to a system (30) for identifying at least one object (5) at least partially immerged in a water area (3), said system (30) comprising a capturing module (310) comprising at least one camera, said at least one camera being configured to generate at least one sequence of images of said water area (3), and a processing module (320) being configured to receive at least one sequence of images from said at least one camera and comprising at least one artificial neural network, said at least one artificial neural network being configured to detect at least one object (5) in said at least one received sequence of images, extract a set of features from said at least one detected object (5), compare said extracted set of features with at least one predetermined set of features associated with a predefined object (5), identify the at least one detected object (5) when the extracted set of features matches with the at least one predetermined set of features.