Multi-Level Feature Extraction for Fast Mobile Object Detection

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image object detection algorithms have low running speeds and high resource requirements, making them unsuitable for implementation on mobile terminals with limited configurations.

Innovation Solution

An image object detection method utilizing an n-level deep feature map framework and an m-level non-deep feature map framework, combined with a feature prediction model, to extract and regress object type and position information efficiently, reducing resource requirements and increasing speed.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing image object detection algorithms are used to achieve accurate detection results, then detection accuracy is improved, but running speed deteriorates and resource consumption increases

Engineering Contradiction:
Improvedetection accuracyVSAvoidrunning speed
Core Design Contradiction:
Measurement precisionVSProductivity

Solution Approach 1:

The patent segments the feature extraction process into multiple levels (first-level, second-level, third-level feature maps) with different abstraction depths. Each level processes features at different scales, allowing the system to balance between detailed feature recognition and processing speed by selectively using features from different levels rather than processing all features uniformly.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces a hierarchical dimension to feature processing by creating multiple levels of feature maps with increasing abstraction. This dimensional transformation allows the system to process features at different granularities simultaneously, improving both accuracy (through deep features) and speed (through shallow features that require less processing).

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

2Measurement precision

If existing image object detection algorithms are used to achieve accurate detection results, then detection accuracy is improved, but resource configuration requirements increase

Engineering Contradiction:
Improvedetection accuracyVSAvoidresource configuration
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent divides the computational workload into segments corresponding to different feature levels. The first-level feature maps handle basic edge and texture detection with lower computational requirements, while second and third-level maps handle more complex pattern recognition. This segmentation allows mobile devices with limited resources to perform object detection by distributing computations across manageable levels rather than requiring all resources to be dedicated to a single complex processing stage.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent applies local quality by using different feature extraction strategies for different parts of the feature hierarchy. Shallow feature maps use simpler extraction methods suitable for basic patterns, while deeper maps employ more sophisticated methods for complex object recognition. This localized approach optimizes resource usage by applying computational intensity only where needed in the feature hierarchy.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3742394B1Image target detection method and apparatus, storage medium, and electronic device
Publication Date: 2025.07.30 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP3742394B1 patent drawingFigure 1~2
  • EP3742394B1 patent drawingFigure 3a~3b
  • EP3742394B1 patent drawingFigure 4a~5a

AI summary

An embodiment of this application provides a method for detecting an object in an image, including: obtaining a detection image, an n-level deep feature map framework and an m-level non-deep feature map framework, the n-level deep feature map framework and the m-level non-deep feature map framework including a feature size and a feature dimension; extracting, based on a deep feature extraction model, a deep feature from an (i-1)-level feature of the detection image using an i-level deep feature map framework, to obtain an i-level feature of the detection image; extracting, based on a non-deep feature extraction model, a non-deep feature from a (j-1+n)-level feature of the detection image using a j-level non-deep feature map framework, to obtain a (j+n)-level feature of the detection image; and performing, based on a feature prediction model, an information regression operation on an a-level feature to an (m+n)-level feature of the detection image, to obtain object type information and object position information of an object in the detection image. An embodiment of this application further provides an apparatus for detecting an object in an image, a storage medium and an electronic device. In the embodiments of this application, an overall speed of feature extraction is relatively high, and a requirement for a configuration resource is relatively low.