Depth-Based Scene Contour Recognition From Multiple Video Images
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing scene contour recognition methods in automation devices require large amounts of data for training, leading to inaccurate recognition due to data dependence and computational inefficiencies.
Innovation Solution
A method that utilizes depth information to determine three-dimensional information of target planes in scene images, generating three-dimensional contours by fusing these planes, and projecting them onto a two-dimensional plane to improve recognition accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep learning models are trained with large quantities of data samples, then recognition capability is improved, but training complexity and data dependence increase leading to inaccurate recognition
Solution Approach 1:
The patent extracts depth information from scene images to generate three-dimensional contours of target objects. By extracting only the essential depth data rather than using complete image datasets for training, the system achieves accurate contour recognition without requiring large quantities of training samples, thus reducing training complexity while maintaining recognition accuracy
Solution Approach 2:
The patent creates a three-dimensional contour representation as a simplified copy of the target object's geometric structure. This 3D contour model serves as a compact representation that captures essential shape information without requiring the complexity of full image datasets, enabling accurate recognition with reduced data requirements
2Measurement precision
If deep learning models are trained with large quantities of data samples, then recognition capability is improved, but the model becomes overly dependent on data samples causing inaccurate recognition
Solution Approach 1:
The patent extracts depth information as a fundamental geometric property from scene images. By building recognition models based on extracted 3D contour structures rather than raw image data, the system achieves reliability that is independent of specific training samples, as the depth-based representations capture universal geometric characteristics of objects
Solution Approach 2:
The patent transforms image data into three-dimensional contour parameters that represent object geometry. This parameter transformation creates a representation space where object recognition is based on geometric properties rather than pixel data, making the recognition system more reliable and less dependent on specific training samples
3Measurement precision
If three-dimensional information is processed from multiple scene images, then contour recognition accuracy is improved, but processing time and computational resources increase
Solution Approach 1:
The patent extracts only the depth information component from scene images to generate three-dimensional contours. By extracting only the essential geometric data rather than processing complete image sequences, the system achieves accurate contour recognition with reduced computational time and resources
Solution Approach 2:
The patent segments the image processing task into distinct stages: extracting depth information, generating three-dimensional contours, and projecting to two-dimensional representations. This segmentation allows each stage to be optimized independently, improving overall processing efficiency while maintaining recognition accuracy
Data Source
AI summary
A scene contour recognition method is provided. In the method, a plurality of scene images of an environment is obtained. Three-dimensional information of a target plane in the plurality of scene images is determined based on depth information for each of the plurality of scene images, The target plane corresponds to a target object in the plurality of scene images. A three-dimensional contour corresponding to the target object is generated by fusing the target plane in each of the plurality of scene images based on the three-dimensional information of the target plane in each of the plurality of scene images. A contour diagram of the target object is generated by projecting the three-dimensional contour onto a two-dimensional plane.