Lightweight Human Pose Estimation Using Segmented Feature Extraction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current human pose estimation methods are resource-intensive and not suitable for real-time performance on resource-limited equipment, especially in multi-person pose estimation scenarios.
Innovation Solution
A lightweight two-dimensional multi-person human pose estimation method using a bottom-up framework that incorporates Part Affinity Fields (PAFs) and heat maps, with a reduced model size and fewer stages to accommodate resource-limited systems.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If current human pose estimation methods are used, then pose estimation accuracy is maintained, but resource consumption is excessively high and real-time performance cannot be achieved on resource-limited equipment
Solution Approach 1:
The pose estimation process is divided into multiple stages (e.g., initial stage, refinement stage) that process information sequentially. Each stage extracts features at different levels of detail, allowing the system to achieve accurate results without processing all data at maximum computational intensity simultaneously, thus enabling real-time performance on resource-limited devices.
Solution Approach 2:
The method performs pose estimation for only the necessary number of people detected in the scene rather than processing entire images at full resolution regardless of content. The system adjusts processing depth based on the number of targets, achieving real-time performance by performing partial processing only where needed.
2Adaptability or versatility
If multi-person pose estimation is performed, then comprehensive pose detection capability is achieved, but computational complexity increases significantly
Solution Approach 1:
The system separates the pose estimation process into distinct stages: initial detection stage that identifies all people and performs coarse pose estimation, and refinement stages that improve accuracy for detected individuals. This segmentation allows multi-person detection without overwhelming computational complexity at any single processing point.
Solution Approach 2:
The computational resources allocated to pose estimation dynamically adjust based on the number of people detected. When fewer people are present, the system can allocate more resources per person for higher accuracy. When many people are detected, the system processes them more efficiently with fewer resources per individual, maintaining overall real-time performance while preserving multi-person detection capability.
Data Source
AI summary
A method of human pose estimation, including, receiving an image frame, extracting in at least a first stage a first stage image feature based on the image frame, extracting in at least a second stage a second stage image feature based on the first stage image feature, extracting in at least a subsequent stage a subsequent stage image feature based on the second stage image feature, up-sampling the subsequent stage image feature, up-sampling the second stage image feature, concatenating the first stage image feature, the up-sampled second stage image feature and the up-sampled subsequent stage image feature and outputting a feature map based on an output of the concatenation.