Pose Identification Using Coordinate Down-Sampling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current pose identification methods using depth images face limitations in precision due to the inability to effectively utilize depth information and require significant memory and computational resources, leading to low-resolution feature maps and decreased accuracy.
Innovation Solution
The proposed method involves obtaining 3D coordinate information corresponding to features in depth images, using a coordinate maintenance module to perform feature and coordinate down-sampling simultaneously, and integrating this information into the pose identification process to enhance precision.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional feature extraction methods are used on depth images, then computational resources and memory are significantly consumed, but the output feature maps become low-resolution leading to decreased pose identification accuracy
Solution Approach 1:
The patent segments the feature extraction process into multiple resolution levels. A down-sampling operation is performed to generate a lower-resolution feature map that requires fewer computational resources, while position information is preserved and integrated to maintain accuracy. This segmentation allows the system to process features at different scales efficiently.
Solution Approach 2:
The patent introduces position information as an additional dimension to the feature extraction process. By incorporating spatial position data alongside feature values, the system can maintain accurate pose identification even when using down-sampled lower-resolution feature maps, thus resolving the trade-off between computational efficiency and accuracy.
2Productivity
If traditional feature extraction methods are used on depth images, then computational resources and memory are significantly consumed, but the pose identification accuracy decreases due to low-resolution feature maps
Solution Approach 1:
The feature extraction process is divided into stages with down-sampling operations that reduce computational load while preserving essential information. The patent processes features at multiple resolution levels rather than maintaining full resolution throughout, improving processing efficiency.
Solution Approach 2:
Position information is added as a critical dimension to the feature representation. This allows the system to achieve accurate pose identification using down-sampled feature maps by incorporating spatial location data, thus maintaining precision while improving productivity.
3Device complexity
If feature down-sampling is performed to reduce computational requirements, then memory and processing resources are reduced, but position information of features is lost
Solution Approach 1:
The patent performs down-sampling operations in a controlled manner while preserving position information. By预先 (in advance) planning the down-sampling strategy to maintain spatial coordinates, the system reduces memory requirements without losing critical position data needed for accurate pose identification.
Solution Approach 2:
Position information is maintained as a separate dimensional component during down-sampling. Rather than losing spatial data when reducing feature map resolution, the patent preserves position information as an additional dimension that can be integrated with the down-sampled features.
Data Source
Figure 1
Figure 2A
Figure 2B
AI summary
Disclosed is a pose identification method including obtaining a depth image of a target, obtaining feature information of the depth image and position information corresponding to the feature information, and obtaining a pose identification result of the target based on the feature information and the position information.