3D Edge Detection Refinement Using Stacked Slice Features

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Current edge detection technologies are limited to two-dimensional images and lack effective methods for accurate three-dimensional edge detection, which is crucial for tasks like semantic segmentation and object tracking.

Innovation Solution

A method involving three-dimensional edge refinement detection network that stacks and encodes two-dimensional object and edge detection results, utilizing a multi-stage encoder and decoder structure with mutual learning modules and atrous convolutions to enhance and optimize edge detection accuracy.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If two-dimensional edge detection methods are used, then the detection process is simple, but the detection accuracy for three-dimensional images is low

Engineering Contradiction:
Improveedge detection accuracyVSAvoiddetection network complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent transitions from two-dimensional edge detection to three-dimensional edge detection by processing multiple two-dimensional slices along the depth dimension. The encoder-decoder network operates on three-dimensional data, stacking multiple two-dimensional detection results and performing three-dimensional convolution to capture spatial relationships in all three dimensions, thereby improving edge detection accuracy for three-dimensional images.

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

Solution Approach 2:

The patent employs a nested encoder-decoder structure where the encoder processes input features through multiple stages with increasing complexity, and the decoder reconstructs the output through corresponding stages. This nested architecture allows the system to progressively extract and refine three-dimensional edge features while maintaining manageable computational complexity at each stage.

Inventive Principle:
Principle #7Nested doll (Nesting)

2Measurement precision

If three-dimensional edge detection is implemented, then the detection accuracy improves, but the computational complexity increases

Engineering Contradiction:
Improvethree-dimensional edge detection accuracyVSAvoidcomputational power consumption
Core Design Contradiction:
Measurement precisionVSPower

Solution Approach 1:

The patent divides three-dimensional image processing into multiple two-dimensional slice operations. By processing each slice independently through the encoder-decoder network and then aggregating results through three-dimensional convolution, the system reduces the computational burden compared to processing the entire three-dimensional volume at once, while still achieving accurate three-dimensional edge detection.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent performs multiple encoding and decoding operations on the same input data to extract features at different scales and representations. This partial repetition of processing stages allows the system to capture comprehensive three-dimensional edge information while distributing computational load across multiple passes rather than requiring a single computationally intensive operation.

Inventive Principle:
Principle #16Partial or excessive action

3Loss of information

If multiple encoding and decoding stages are used, then the feature extraction is enhanced, but the processing time increases

Engineering Contradiction:
Improvefeature information retentionVSAvoidprocessing time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The patent implements continuous feature extraction through the encoder-decoder network where each stage builds upon the previous stage's output without interruption. The multi-stage encoding progressively extracts features at different levels of abstraction, and the corresponding decoding stages continuously reconstruct the feature maps, ensuring that useful information is preserved and refined throughout the entire processing pipeline without loss of critical edge features.

Inventive Principle:
Principle #20Continuity of useful action

Data Source

PatentEP4016454B1Three-dimensional edge detection method and apparatus, storage medium and computer device
Publication Date: 2025.09.03 TENCENT TECHNOLOGY (SHENZHEN) CO LTD
  • EP4016454B1 patent drawingFigure 1~2
  • EP4016454B1 patent drawingFigure 3~4
  • EP4016454B1 patent drawingFigure 5

AI summary

A three-dimensional edge detection method and apparatus, a computer-readable storage medium and a computer device. Said method comprises: acquiring two-dimensional object detection results and two-dimensional edge detection results of two-dimensional fragments of a three-dimensional image; stacking the two-dimensional object detection results into a three-dimensional object detection result, and stacking the two-dimensional edge detection results into a three-dimensional edge detection result; performing encoding according to a feature map of the three-dimensional image, the three-dimensional object detection result and the three-dimensional edge detection result, so as to obtain an encoded result; and performing decoding according to the encoded result, the three-dimensional object detection result and the three-dimensional edge detection result, so as to obtain an optimized three-dimensional edge detection result of the three-dimensional image.