Scene Recognition via Object Detection Model

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Conventional scene recognition methods are limited to recognizing fixed classes of scenes and require manual updates for new scene requirements, leading to high costs, low efficiency, and low precision.

Innovation Solution

A scene recognition method that uses an object detection model comprising a backbone network, feature pyramid network, convolutional network, and prediction network to recognize scenes by inputting images and object images, allowing for quick and precise recognition of any scene without manual intervention.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If manual scene recognition is performed for new scene requirements, then scene recognition precision can be maintained, but recognition efficiency decreases and costs increase

Engineering Contradiction:
Improvescene recognition precisionVSAvoidscene recognition efficiency
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The object detection model performs automatic scene recognition without requiring manual intervention. The system extracts features from images, detects objects automatically, and determines scene types through algorithmic processing, enabling the system to serve itself rather than requiring human operators for each recognition task.

Inventive Principle:
Principle #25Self-service

Solution Approach 2:

The patent replaces manual mechanical scene recognition with an automated computer vision system. The object detection model uses neural networks and algorithmic processing to substitute human operators, achieving both high precision and improved efficiency through automated image analysis and object detection.

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

2Device complexity

If fixed-class scene recognition is implemented, then system complexity remains low, but adaptability to new scenes deteriorates

Engineering Contradiction:
Improvesystem complexityVSAvoidscene adaptability
Core Design Contradiction:
Device complexityVSAdaptability or versatility

Solution Approach 1:

The object detection model serves multiple functions: it detects various objects (cones, barrels, signboards, vehicles), identifies different scene types (construction, traffic, accident, normal), and adapts to new scene requirements. This universal model replaces the need for separate fixed-class recognition systems, achieving high adaptability without proportionally increasing complexity.

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

Solution Approach 2:

The system transitions from static fixed-class recognition to dynamic scene understanding. The object detection model can identify new scene types by detecting combinations of objects and their spatial relationships, allowing the system to adapt dynamically to new scenarios without requiring pre-programmed categories for every possible scene.

Inventive Principle:
Principle #15Dynamics

3Measurement precision

If manual scene recognition is performed, then recognition precision can be maintained, but time consumption increases

Engineering Contradiction:
Improvescene recognition precisionVSAvoidtime consumption
Core Design Contradiction:
Measurement precisionVSLoss of time

Solution Approach 1:

The object detection model processes images continuously and automatically, maintaining high recognition precision through consistent algorithmic application. The system performs real-time object detection and scene classification without interruption or manual review, eliminating the time loss associated with manual recognition while preserving accuracy through automated feature extraction and analysis.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4451232A1Scene recognition method, method for obtaining scene data, device, medium, and vehicle
Publication Date: 2024.10.23 ANHUI NIO AUTONOMOUS DRIVING TECH CO LTD
  • EP4451232A1 patent drawingFigure 1~3
  • EP4451232A1 patent drawingFigure 4~6
  • EP4451232A1 patent drawing

AI summary

The disclosure relates to the field of image recognition technologies, and specifically provides a scene recognition method, a method for obtaining scene data, a device, a medium, and a vehicle, to resolve technical problems of low scene recognition efficiency and low recognition precision of an existing scene recognition method. The scene recognition method of the disclosure includes: obtaining an image to be recognized and at least one object image; constructing and training an object detection model; and recognizing a current scene based on the object detection model, the image to be recognized, and the at least one object image, to obtain a scene recognition result. In this way, any scene can be quickly recognized, and a scene recognition result with high precision can be obtained.