Human Body Depth Mapping from Skeletal Points Without Deep Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for generating human body depth maps using deep learning models are resource-intensive and time-consuming, leading to high costs and low efficiency.
Innovation Solution
Generate human body depth maps by acquiring skeletal point positions in a two-dimensional image coordinate system and distances to a camera imaging plane in a three-dimensional coordinate system, determining depth values based on these positions and distances, and drawing on a blank canvas to create the depth map.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a deep learning model is used to generate human body depth maps, then the quality and accuracy of depth maps can be improved, but hardware resource requirements and time consumption increase significantly
Solution Approach 1:
The patent segments the human body into multiple skeletal parts (trunk, head, upper limbs, lower limbs) and processes each part separately using template matching. This divides the complex full-body depth estimation problem into smaller, more efficient sub-problems that can be solved without requiring heavy deep learning models.
Solution Approach 2:
The patent extracts skeletal information from the input image and uses it as a guide for depth map generation. By focusing on the skeletal structure rather than processing the entire image through a deep learning model, the method reduces computational complexity while maintaining accuracy.
2Measurement precision
If a deep learning model is used to generate human body depth maps, then the quality of depth maps can be improved, but hardware resource occupation and costs increase
Solution Approach 1:
The patent uses pre-defined skeletal templates that represent typical human body structures. These templates are copied and adapted to match the detected skeletal points in the input image, avoiding the need for complex deep learning models while maintaining reasonable accuracy for standard poses.
Solution Approach 2:
The patent employs lightweight template matching algorithms instead of resource-intensive deep learning models. This substitutes expensive computational resources with simpler, faster methods that consume fewer hardware resources.
3Measurement precision
If a deep learning model is used to generate human body depth maps, then the accuracy can be improved, but time consumption increases
Solution Approach 1:
The patent performs skeletal point detection and template preparation in advance. By pre-processing the skeletal structure identification and having templates ready, the actual depth map generation becomes much faster, as the computationally intensive parts are done beforehand.
Solution Approach 2:
The patent skips the time-consuming deep learning inference process by using direct template matching on detected skeletal points. This allows the method to rapidly generate depth maps without waiting for iterative model processing.
Data Source
AI summary
A method for generating a human body depth map includes: acquiring a skeletal point position of each skeletal point in a skeleton of a target object in a two-dimensional image coordinate system, and acquiring a distance from each skeletal point to a camera imaging plane in a three-dimensional camera coordinate system; acquiring a blank canvas for drawing a human body depth map; determining a depth value of each first pixel point of the skeleton based on the skeletal point position and the distance of each skeletal point; and drawing in the blank canvas based on the depth value of each first pixel point, to obtain a human body depth map of the target object.